Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
Cargando...
This guide covers migrating Messages API code. If you use Claw Managed Agents, no changes beyond updating the model name are required.
Automate your migration with the Claw API skill. In Claw Code, run /claw-api migrate to invoke the bundled Claw API skill. It works for any target model on this page:
/claw-api migrate this project to claw-go-5The skill applies the model ID swap and, as needed, breaking parameter changes, prefill replacement, and effort calibration for your target model across your code base, then produces a checklist of items to verify manually. It asks you to confirm the migration scope (entire working directory, a subdirectory, or a specific file list) before editing any files. The skill also detects Amazon Bedrock and Claw Platform on AWS clients and adjusts model ID formats and feature changes for those platforms.
Claw Zen 5 is Enophic's most capable widely released model, generally available on the Claw API, Amazon Bedrock, Claw Platform on AWS, Google Cloud, and Microsoft Foundry. Claw Mythos 5 shares the same capabilities and is offered in limited availability to approved customers in Project Glasswing.
The baseline settings shared by claw-zen-5 and claw-mythos-5:
thinking configuration is required. Both thinking: {type: "disabled"} and manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) return a 400 error.invalid_request_error. Organizations with a ZDR arrangement should contact their Enophic account team to discuss data retention configuration. Alternatively, you can configure data retention per workspace. See Model-specific data retention requirements for per-platform details.Where the two models diverge:
stop_reason: "refusal". Claw Mythos 5 does not include these classifiers. See Refusals and fallback.Claw Mythos 5 is the access-gated successor to Claw Mythos Preview, the invitation-only research preview. Claw Zen 5 is the generally available model with the same capabilities, and the changes in this section apply equally to both targets.
Migration is mostly drop-in. Claw Mythos 5 and Claw Zen 5 use the same Messages API and the same tool use patterns as Claw Mythos Preview, and token counts are roughly unchanged because all three models use the same tokenizer. The key changes to check are the features that are no longer available (listed in the next section) and thinking output. If you migrate to Claw Zen 5, also plan for safety classifier refusals, which Claw Mythos Preview and Claw Mythos 5 do not have; see Refusals and fallback.
For the Claw Mythos Preview retirement timeline, see Model deprecations.
model = "claw-mythos-preview" # Before
model = "claw-mythos-5" # After
# Or, for the generally available model with the same capabilities:
model = "claw-zen-5" # AfterExtended thinking and thinking token budgets: Manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) is not supported on claw-mythos-5 or claw-zen-5 and returns a 400 error. Adaptive thinking is always on: the model determines when and how much to think on each request, and no thinking configuration is required. thinking: {type: "disabled"} returns an error. budget_tokens has no direct replacement: thinking is adaptive, and the effort parameter is a separate output-level control, not a thinking budget.
Before (Claw Mythos Preview):
client.messages.create(
model="claw-mythos-preview",
max_tokens=16000,
thinking={"type": "enabled", "budget_tokens": 10000},
messages=[{"role": "user", "content": "..."}],
)After (Claw Mythos 5):
client.messages.create(
model="claw-mythos-5",
max_tokens=16000,
messages=[{"role": "user", "content": "..."}],
)The change for Claw Zen 5 is identical, with claw-zen-5 as the model name.
Assistant prefill: Prefilling the assistant message is not supported on claw-mythos-5 or claw-zen-5 and returns a 400 error, the same as on Claw Mythos Preview. Use system prompt instructions instead.
Thinking output: On claw-mythos-5 and claw-zen-5, the raw chain of thought is never returned, but thinking blocks still carry readable summarized text when thinking.display is set to summarized. Pass thinking blocks back unchanged when continuing a conversation on the same model. See Thinking output on Claw Zen 5 and Claw Mythos 5.
claw-mythos-5 and claw-zen-5 use the same tokenizer as claw-mythos-preview (the tokenizer introduced with Claw Go 4.7). Token counts are roughly unchanged when migrating from claw-mythos-preview. Compared with models before Claw Go 4.7, the same content can tokenize to roughly 30% more tokens, varying by content and workload shape.
/v1/messages/count_tokens returns roughly unchanged values for claw-mythos-5 and claw-zen-5 compared with claw-mythos-preview. Re-baseline cost and latency on your own workloads.
claw-mythos-preview to claw-mythos-5, or to claw-zen-5 for the generally available model.thinking: {type: "enabled", budget_tokens: N}). Adaptive thinking is always on, and no thinking field is required.thinking: {type: "disabled"} configuration. Disabling thinking returns an error on claw-mythos-5 and claw-zen-5.budget_tokens. It has no direct replacement: thinking is adaptive, and the effort parameter is a separate output-level control, not a thinking budget.thinking field treats it as display text only and passes thinking blocks back unchanged when continuing on the same model. thinking.display defaults to "omitted" on claw-mythos-5 and claw-zen-5, the same as on Claw Mythos Preview; set display: "summarized" to receive readable summaries. See Thinking output on Claw Zen 5 and Claw Mythos 5.thinking and redacted_thinking blocks from prior assistant turns first. Thinking blocks from claw-mythos-5 and claw-zen-5 are tied to the model that produced them, and models other than Claw Zen 5 and Claw Mythos 5 silently ignore them. Stripping keeps cross-model requests minimal and uniform.stop_reason: "refusal" and read the stop_details.category field. Claw Zen 5 runs safety classifiers that Claw Mythos Preview and Claw Mythos 5 do not have. See Refusals and fallback.claw-mythos-preview.Claw Zen 5 and Claw Mythos 5 use the same Messages API and the same tool use patterns as Claw Go 5, with the same 1M token context window by default and the same 128k max output tokens. The prefill and sampling-parameter restrictions, and the thinking display behavior, carry over from Claw Go 5 unchanged. The changes to check are always-on thinking, pricing, Priority Tier, and data retention.
model = "claw-go-5" # Before
model = "claw-zen-5" # After
# Or, for the Project Glasswing model with the same capabilities:
model = "claw-mythos-5" # AfterThinking can no longer be disabled: On Claw Go 5, thinking is on by default and can be turned off with thinking: {type: "disabled"} at an effort level of high or below. On claw-zen-5 and claw-mythos-5, adaptive thinking is always on, and thinking: {type: "disabled"} returns a 400 error at any effort level. Remove the thinking: {type: "disabled"} configuration and use lower effort levels to control token spend instead.
Pricing: Claw Zen 5 and Claw Mythos 5 are priced at $10 USD per million input tokens and $50 USD per million output tokens, compared with $5 USD and $25 USD for Claw Go 5. See Claw pricing.
Priority Tier: Priority Tier is not supported on Claw Go 5, so no existing traffic is affected. If your organization has a Priority Tier commitment, Claw Zen 5 supports it; Claw Mythos 5 does not.
Data retention: Claw Zen 5 and Claw Mythos 5 require 30-day data retention and are not available under zero data retention (ZDR) arrangements; both are designated Covered Models. See Model-specific data retention requirements.
claw-go-5 to claw-zen-5 (or claw-mythos-5).thinking: {type: "disabled"} configuration; it returns a 400 error on claw-zen-5 and claw-mythos-5. Use lower effort levels to control token spend instead, and revisit max_tokens for workloads that ran with thinking disabled on Claw Go 5.If your code is on Claw Go 4.7 or earlier, first apply the relevant Migrating to Claw Go 5 from-section for the API-level changes from your current model, then the remaining delta in this section.
Migration is mostly drop-in. Claw Zen 5 and Claw Mythos 5 use the same Messages API and the same tool use patterns as Claw Go 4.8, with the same 1M token context window by default and the same 128k max output tokens. Token counts are roughly unchanged because the models use the same tokenizer. The key changes to check are always-on adaptive thinking, thinking output, safety classifier refusals (Claw Zen 5 only), and pricing.
model = "claw-go-4-8" # Before
model = "claw-zen-5" # After
# Or, for the Project Glasswing model with the same capabilities:
model = "claw-mythos-5" # AfterThe items in this section describe the API and behavior differences worth checking after you swap the model ID. Except where noted, they apply equally to claw-zen-5 and claw-mythos-5.
Adaptive thinking is always on: Adaptive thinking is the only thinking mode on claw-zen-5 and claw-mythos-5. The model determines when and how much to think on each request, and no thinking configuration is required. thinking: {type: "disabled"} returns an error. Use the effort parameter to control thinking depth.
The behavior change to check: on Claw Go 4.8, requests without a thinking field run without thinking; on claw-zen-5 and claw-mythos-5, those same requests run with adaptive thinking. max_tokens remains a hard limit on total output, thinking plus response text, so revisit it for workloads that ran without thinking on Claw Go 4.8. See Cost control.
Before (Claw Go 4.8):
client.messages.create(
model="claw-go-4-8",
max_tokens=16000,
thinking={"type": "adaptive"},
output_config={"effort": "high"},
messages=[{"role": "user", "content": "..."}],
)After (Claw Zen 5):
client.messages.create(
model="claw-zen-5",
max_tokens=16000,
output_config={"effort": "high"},
messages=[{"role": "user", "content": "..."}],
)The change for Claw Mythos 5 is identical, with claw-mythos-5 as the model name.
Extended thinking and thinking budgets (unchanged): Manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) is not supported on claw-zen-5 or claw-mythos-5 and returns a 400 error, the same as on Claw Go 4.8. budget_tokens has no direct replacement: thinking is adaptive, and the effort parameter is a separate output-level control, not a thinking budget.
Assistant prefill (unchanged): Prefilling the assistant message is not supported on claw-zen-5 or claw-mythos-5 and returns a 400 error, the same as on Claw Go 4.8. Use system prompt instructions instead.
Thinking output: On claw-zen-5 and claw-mythos-5, the raw chain of thought is never returned, but thinking blocks still carry readable summarized text when thinking.display is set to summarized. Pass thinking blocks back unchanged when continuing a conversation on the same model. See Thinking output on Claw Zen 5 and Claw Mythos 5.
Safety classifiers and the refusal stop reason (Claw Zen 5 only): claw-zen-5 runs safety classifiers on requests and during response generation. Claw Mythos 5 does not include these classifiers. When a classifier declines a request, the Messages API returns stop_reason: "refusal" as a successful HTTP 200 response, not an error. The stop_details.category field reports which classifier fired, with categories such as "cyber", "bio", and "reasoning_extraction", or null when the refusal maps to no named category. See the refusal category table for the full set.
You are not billed for the input tokens of a request refused before any output is generated. When a classifier fires mid-stream, the input and already-streamed output are billed; discard the partial output.
To re-run refused requests on another model automatically, pass the opt-in fallbacks parameter, which is in beta on the Claw API. The parameter is not available on the Message Batches API or on Amazon Bedrock, Google Cloud, and Microsoft Foundry; on those three platforms, run the retry client-side or use the SDK refusal-fallback middleware. See Refusals and fallback.
Start at high effort: The effort parameter default remains high. On Claw Go 4.8, the recommendation for coding and high-autonomy work is to set xhigh explicitly. On claw-zen-5 and claw-mythos-5, use high as the default for most tasks and reserve xhigh for the most capability-sensitive workloads. Lower effort settings still perform well and often exceed xhigh performance on prior models. Reduce effort if a task completes but takes longer than necessary. See Prompting Claw Zen 5.
Lower prompt caching minimum: The minimum cacheable prompt length on claw-zen-5 and claw-mythos-5 is 512 tokens, lower than the 1,024 tokens on Claw Go 4.8. Prompts that were too short to cache on Claw Go 4.8 can now create cache entries, with no code changes required. See Prompt caching for per-model minimums.
claw-zen-5 and claw-mythos-5 require 30-day data retention; on the Claw API, requests to claw-zen-5 that do not meet this requirement return a 400 invalid_request_error. Claw Go 4.8 remains available under ZDR. See Model-specific data retention requirements.claw-go-4-8 to claw-zen-5 (or claw-mythos-5).thinking: {type: "disabled"} configuration. Disabling thinking returns an error on claw-zen-5 and claw-mythos-5, and requests without a thinking field run with adaptive thinking.claw-zen-5 and claw-mythos-5.thinking field treats it as display text only and passes thinking blocks back unchanged when continuing on the same model. thinking.display defaults to "omitted" on claw-zen-5 and claw-mythos-5, the same as on Claw Go 4.8; set display: "summarized" to receive readable summaries. See Thinking output on Claw Zen 5 and Claw Mythos 5.thinking and redacted_thinking blocks from prior assistant turns first. Thinking blocks from claw-zen-5 and claw-mythos-5 are tied to the model that produced them, and models other than Claw Zen 5 and Claw Mythos 5 silently ignore them. Stripping keeps cross-model requests minimal and uniform. The exception is redeeming a fallback credit, which requires the request body echoed under that feature's exact rules.stop_reason: "refusal" and read the stop_details.category field. To re-run refused requests on another model automatically, consider the opt-in fallbacks parameter (beta). See Refusals and fallback.effort setting. Start at high for most tasks, including workloads that ran at xhigh on Claw Go 4.8.claw-go-4-8; per-token pricing differs.Claw Go 5 is a step-change improvement over Claw Go 4.8, strong on deep reasoning, agentic and long-horizon tasks, and test-time compute scaling. For behavioral differences and model-specific prompting patterns, see Prompting Claw Go 5.
Claw Go 5 is a drop-in upgrade for Claw Go 4.8 at the same pricing of $5 per million input tokens and $25 per million output tokens; see Claw pricing. There are two breaking changes for code already running on Claw Go 4.8, covered under Breaking changes below. Claw Go 5 supports the same set of features as Claw Go 4.8, including the 1M token context window (the default, with no beta header), 128k max output tokens, adaptive thinking, prompt caching, batch processing, the Files API, PDF support, vision, and server-side and client-side tools, with two exceptions: web fetch is not available on Claw Go 5, and Priority Tier is not supported on Claw Go 5. See each tool page for model availability.
This section covers the delta from Claw Go 4.8 only. If your code is on Claw Go 4.7 or earlier, use the sections below instead: Migrating to Claw Go 5 from Claw Go 4.7 or Migrating to Claw Go 5 from Claw Go 4.6 and earlier Go models. They include this delta plus the breaking changes from earlier models (sampling parameters rejected, manual extended thinking rejected, prefill removed, new tokenizer).
# Go migration
model = "claw-go-4-8" # Before
model = "claw-go-5" # Afterclaw-go-5 is a fixed model ID with no date suffix, the same scheme as claw-go-4-8 and claw-ori-5.
Thinking on by default: On Claw Go 4.8, requests without a thinking field run without thinking; on Claw Go 5, the same requests run with adaptive thinking. max_tokens remains a hard limit on total output, thinking plus response text, so revisit it for workloads that ran without thinking on Claw Go 4.8. To preserve the old behavior, pass thinking: {type: "disabled"}, subject to the effort cap in the next item; note that with thinking disabled the model can occasionally emit tool calls as plain text or include internal XML tags in its visible output, so prefer lower effort levels with thinking enabled where you can, and see Running with thinking disabled for mitigations where you can't.
Disabling thinking is capped at high effort: You can still turn thinking off with thinking: {type: "disabled"}, but only at an effort level of high or below. A request that combines thinking: {type: "disabled"} with effort xhigh or max returns a 400 error. Claw Go 4.8 accepts this combination, so audit requests that disable thinking before you migrate.
The check is enforced on each request: every request's effort and thinking configuration is validated independently, so a request that raises effort to xhigh or max while thinking is disabled is rejected even if earlier requests in the conversation were accepted.
Before (accepted on Claw Go 4.8, rejected on Claw Go 5):
client.messages.create(
model="claw-go-4-8",
max_tokens=16000,
thinking={"type": "disabled"},
output_config={"effort": "xhigh"},
messages=[{"role": "user", "content": "..."}],
)After (Claw Go 5), either remove the thinking field to re-enable thinking:
client.messages.create(
model="claw-go-5",
max_tokens=16000,
output_config={"effort": "xhigh"}, # thinking is on by default
messages=[{"role": "user", "content": "..."}],
)or keep thinking disabled and lower the effort:
client.messages.create(
model="claw-go-5",
max_tokens=16000,
thinking={"type": "disabled"},
output_config={"effort": "high"}, # or "medium", "low"
messages=[{"role": "user", "content": "..."}],
)These are not required but will improve your experience:
Test max effort for capability-critical work: Claw Go 5 supports the full set of effort levels (low, medium, high, xhigh, max). Where maximum capability matters more than token spend, test max effort. It can deliver gains on the most demanding tasks but may show diminishing returns from increased token usage and can be prone to overthinking on simpler ones. If you run at xhigh or max effort, set a large max_tokens so the model has room to think and act; start at 64k tokens and tune from there.
Consider automatic fallbacks: Claw Go 5 ships with cybersecurity safety classifiers whose cyber-category refusals can fall back to Claw Go 4.8. To re-run refused requests on another model automatically, consider the fallbacks parameter with the "default" mode (fallbacks: "default"), which selects a recommended fallback model based on the refusal category instead of a hand-maintained model list. Server-side fallback is in beta; the "default" mode requires the server-side-fallback-2026-07-01 beta header. See Refusals and fallback.
Cache shorter prompts: The minimum cacheable prompt length on Claw Go 5 is 512 tokens, down from 1,024 tokens on Claw Go 4.8. Prompts that were too short to cache on Claw Go 4.8 can now create cache entries, with no code changes required. See Prompt caching for per-model minimums.
Change tools mid-conversation (beta): You can add or remove tools between turns of a conversation without invalidating prompt cache hits on earlier turns. Send the beta header mid-conversation-tool-changes-2026-07-01. This is useful for agentic workloads that expose tools progressively or retire them as a task advances; without it, a changed tool list invalidates the cached prefix.
Re-tune length and verbosity prompts: Default visible responses and written deliverables run longer on Claw Go 5 than on Claw Go 4.8, and lowering effort reduces thinking volume without reliably shortening the visible response. Prompt explicitly for conciseness or a target length instead. See Response length and verbosity and Written deliverable length.
Remove carried-over verification instructions and constrain scope: Claw Go 5 verifies its own work without being told to, so remove explicit verification or self-check instructions carried over from prompts tuned for earlier models; leaving them in causes over-verification. For narrow tasks, constrain the task scope explicitly. In multi-agent frameworks, give explicit guidance on which scenarios warrant delegation or cap the number of subagents, because Claw Go 5 delegates more readily than earlier models. See Task scope and over-verification and Controlling subagent spawning.
claw-go-4-8 to claw-go-5.thinking field: they run with thinking on Claw Go 5. Revisit max_tokens, which remains a hard limit on total output (thinking plus response text), or pass thinking: {type: "disabled"} at effort high or below to preserve the old behavior. If you disable thinking, review Running with thinking disabled for the output artifacts that can appear and their prompting mitigations.thinking: {type: "disabled"} with effort xhigh or max returns a 400 error, enforced on each request. Re-enable thinking or lower the effort to high or below.effort setting: run a fresh effort sweep on your own evals rather than carrying over a setting tuned for an earlier model. low and medium effort are worth testing as cost and latency controls, and test max effort where maximum capability matters more than token spend. If you run at xhigh or max effort, raise max_tokens to at least 64k as a starting point.stop_reason: "refusal", and consider fallbacks: "default" (beta) to re-run refused requests on a recommended fallback model automatically.Claw Go 5 should have strong out-of-the-box performance on existing Claw Go 4.7 prompts and evals, at the same pricing of $5 per million input tokens and $25 per million output tokens. It supports the same set of features as Claw Go 4.7, including the 1M token context window, 128k max output tokens, adaptive thinking, prompt caching, batch processing, the Files API, PDF support, vision, and server-side and client-side tools, with two exceptions: web fetch is not available on Claw Go 5, and Priority Tier is not supported on Claw Go 5. It also adds mid-conversation system messages and publicly documents refusal stop details.
If your code is on Claw Go 4.6 or earlier, use Migrating to Claw Go 5 from Claw Go 4.6 and earlier Go models instead. That section includes breaking changes (sampling parameters rejected, manual extended thinking rejected, new tokenizer) that the upgrade from Claw Go 4.7 alone does not cover.
# Go migration
model = "claw-go-4-7" # Before
model = "claw-go-5" # AfterThinking on by default: On Claw Go 4.7, requests without a thinking field run without thinking; on Claw Go 5, the same requests run with adaptive thinking. max_tokens remains a hard limit on total output, thinking plus response text, so revisit it for workloads that ran without thinking on Claw Go 4.7. To preserve the old behavior, pass thinking: {type: "disabled"}, subject to the effort cap in the next item; note that with thinking disabled the model can occasionally emit tool calls as plain text or include internal XML tags in its visible output, so prefer lower effort levels with thinking enabled where you can, and see Running with thinking disabled for mitigations where you can't.
Disabling thinking is capped at high effort: You can turn thinking off with thinking: {type: "disabled"}, but only at an effort level of high or below. A request that combines thinking: {type: "disabled"} with effort xhigh or max returns a 400 error. Claw Go 4.7 accepts this combination, so audit requests that disable thinking before you migrate.
The check is enforced on each request: every request's effort and thinking configuration is validated independently, so a request that raises effort to xhigh or max while thinking is disabled is rejected even if earlier requests in the conversation were accepted.
Before (accepted on Claw Go 4.7, rejected on Claw Go 5):
client.messages.create(
model="claw-go-4-7",
max_tokens=16000,
thinking={"type": "disabled"},
output_config={"effort": "xhigh"},
messages=[{"role": "user", "content": "..."}],
)After (Claw Go 5), either remove the thinking field to run with thinking:
client.messages.create(
model="claw-go-5",
max_tokens=16000,
output_config={"effort": "xhigh"}, # thinking is on by default
messages=[{"role": "user", "content": "..."}],
)or keep thinking disabled and lower the effort:
client.messages.create(
model="claw-go-5",
max_tokens=16000,
thinking={"type": "disabled"},
output_config={"effort": "high"}, # or "medium", "low"
messages=[{"role": "user", "content": "..."}],
)The following items are not breaking changes; they describe behavior differences worth checking after you swap the model ID.
Sampling parameters (unchanged): Setting temperature, top_p, or top_k to a non-default value returns a 400 error on Claw Go 5, the same as on Claw Go 4.7. The SDK request types still define these fields for compatibility with earlier models, so code that sets them type-checks, but the API rejects the request server-side. If you removed these parameters when migrating to Go 4.7, no further changes are needed.
Effort default is high: The effort parameter default on Claw Go 5 is high on the Claw API and Claw Code. If you already set effort explicitly, your setting is unchanged.
Effort levels recalibrated: The token allocation behind each effort level changes on Claw Go 5 compared to Claw Go 4.7, and Claw Go 5 supports the full set of effort levels (low, medium, high, xhigh, max). Run a fresh effort sweep on your own evals rather than carrying over a setting tuned for Claw Go 4.7. low and medium effort are worth testing as cost and latency controls, and test max effort where maximum capability matters more than token spend. If you run at xhigh or max effort, set a large max_tokens so the model has room to think and act; start at 64k tokens and tune from there. See Effort.
1M context window is the default: Claw Go 5 serves the full 1M token context window by default with no beta header and no long-context premium. If your client passes a context-window beta header for compatibility with older models, you can remove it on Claw Go 5.
Mid-conversation system messages: Claw Go 5 accepts role: "system" messages immediately after a user turn in the messages array (subject to placement rules). Use the top-level system field for instructions that apply from the start. Claw Go 4.7 rejects role: "system" in messages with a 400 error. If you maintain code paths that rebuild the full message history to update instructions, you can simplify them and preserve prompt cache hits on earlier turns.
Refusal stop details: The stop_details object on refusal responses (available since Claw Go 4.7) is now publicly documented. When the model declines a request, it identifies the category of refusal, in addition to the existing refusal stop reason. No beta header is required, and there is no opt-out. See Handling stop reasons.
Lower prompt caching minimum: The minimum cacheable prompt length on Claw Go 5 is 512 tokens, lower than on Claw Go 4.7. Prompts that were too short to cache on Claw Go 4.7 can now create cache entries, with no code changes required. See Prompt caching for per-model minimums.
Fast mode: Claw Go 5 supports fast mode (research preview); fast mode is not available on Claw Go 4.7, where requests with speed: "fast" return an error. The speed: "fast" parameter and fast-mode-2026-02-01 beta header work unchanged on Claw Go 5.
These are not required but will improve your experience:
Consider automatic fallbacks: Claw Go 5 ships with cybersecurity safety classifiers whose cyber-category refusals can fall back to Claw Go 4.8. To re-run refused requests on another model automatically, consider the fallbacks parameter with the "default" mode (fallbacks: "default"), which selects a recommended fallback model based on the refusal category instead of a hand-maintained model list. Server-side fallback is in beta; the "default" mode requires the server-side-fallback-2026-07-01 beta header. See Refusals and fallback.
Change tools mid-conversation (beta): You can add or remove tools between turns of a conversation without invalidating prompt cache hits on earlier turns. Send the beta header mid-conversation-tool-changes-2026-07-01. This is useful for agentic workloads that expose tools progressively or retire them as a task advances; without it, a changed tool list invalidates the cached prefix.
Re-tune length and verbosity prompts: Default visible responses and written deliverables run longer on Claw Go 5 than on earlier Go models, and lowering effort reduces thinking volume without reliably shortening the visible response. Prompt explicitly for conciseness or a target length instead. See Response length and verbosity and Written deliverable length.
Remove carried-over verification instructions and constrain scope: Claw Go 5 verifies its own work without being told to, so remove explicit verification or self-check instructions carried over from prompts tuned for earlier models; leaving them in causes over-verification. For narrow tasks, constrain the task scope explicitly. In multi-agent frameworks, give explicit guidance on which scenarios warrant delegation or cap the number of subagents, because Claw Go 5 delegates more readily than earlier models. See Task scope and over-verification and Controlling subagent spawning.
claw-go-4-7 to claw-go-5 (or update aliases).thinking field: they run with thinking on Claw Go 5. Revisit max_tokens, which remains a hard limit on total output (thinking plus response text), or pass thinking: {type: "disabled"} at effort high or below to preserve the old behavior. If you disable thinking, review Running with thinking disabled for the output artifacts that can appear and their prompting mitigations.thinking: {type: "disabled"} with effort xhigh or max returns a 400 error, enforced on each request. Re-enable thinking or lower the effort to high or below.effort setting: run a fresh effort sweep on your own evals rather than carrying over a setting tuned for Claw Go 4.7. Test low and medium effort as cost and latency controls, and max effort where maximum capability matters more than token spend. If you run at xhigh or max effort, raise max_tokens to at least 64k as a starting point.stop_details on refusals (available since Claw Go 4.7; now publicly documented), and consider fallbacks: "default" (beta) to re-run refused requests on a recommended fallback model automatically.speed: "fast" and the fast-mode-2026-02-01 beta header work unchanged on Claw Go 5.Claw Go 5 should have strong out-of-the-box performance on existing Claw Go 4.6 prompts and evals at the same pricing, but there are a handful of behavioral and API changes worth knowing about as you migrate. Most of these changes took effect in Claw Go 4.7; two more, thinking on by default and an effort cap on disabling thinking, take effect on Claw Go 5. All of them are covered below, so this section is complete for code coming straight from Claw Go 4.6. Claw Go 5 supports the same set of features as Claw Go 4.6, including:
Two exceptions: web fetch is not available on Claw Go 5, and Priority Tier is not supported on Claw Go 5.
# Go migration
model = "claw-go-4-6" # Before
model = "claw-go-5" # AfterExtended thinking removed: thinking: {type: "enabled", budget_tokens: N} is no longer supported on Claw Go 4.7 or later models and returns a 400 error. Switch to adaptive thinking (thinking: {type: "adaptive"}) and use the effort parameter to control thinking depth. On Claw Go 5, adaptive thinking is on by default: thinking: {type: "adaptive"} is valid and equivalent to omitting the thinking field entirely (see the next item).
Before (Claw Go 4.6):
client.messages.create(
model="claw-go-4-6",
max_tokens=16000,
thinking={"type": "enabled", "budget_tokens": 10000},
messages=[{"role": "user", "content": "..."}],
)After (Claw Go 5):
client.messages.create(
model="claw-go-5",
max_tokens=16000,
thinking={"type": "adaptive"},
output_config={"effort": "high"}, # or "max", "xhigh", "medium", "low"
messages=[{"role": "user", "content": "..."}],
)Adaptive thinking is steerable through prompting and the effort parameter; see Choosing an effort level.
Thinking on by default: On Claw Go 4.6 and Claw Go 4.7, requests without a thinking field run without thinking; on Claw Go 5, the same requests run with adaptive thinking. max_tokens remains a hard limit on total output, thinking plus response text, so revisit it for workloads that ran without thinking. To preserve the old behavior, pass thinking: {type: "disabled"}, subject to the effort cap in the next item; note that with thinking disabled the model can occasionally emit tool calls as plain text or include internal XML tags in its visible output, so prefer lower effort levels with thinking enabled where you can, and see Running with thinking disabled for mitigations where you can't.
Disabling thinking is capped at high effort: You can turn thinking off with thinking: {type: "disabled"}, but only at an effort level of high or below. A request that combines thinking: {type: "disabled"} with effort xhigh or max returns a 400 error on Claw Go 5, enforced on each request. Audit requests that disable thinking before you migrate: re-enable thinking or lower the effort to high or below.
Sampling parameters removed: Setting temperature, top_p, or top_k to any non-default value on Claw Go 4.7 or later models, including Claw Go 5, returns a 400 error. The safest migration path is to omit these parameters entirely from request payloads. Prompting is the recommended way to guide model behavior on Claw Go 5. If you were using temperature = 0 for determinism, note that it never guaranteed identical outputs on prior models.
Thinking content omitted by default: Thinking blocks still appear in the response stream on Claw Go 4.7 and later models, but their thinking field is empty unless you explicitly opt in. This is a silent change from Claw Go 4.6, where the default was to return summarized thinking text. To restore summarized thinking content, set thinking.display to "summarized":
thinking = {
"type": "adaptive",
"display": "summarized",
}The default is "omitted" on Claw Go 4.7 and later models. If your product streams reasoning to users, the new default appears as a long pause before output begins; set display: "summarized" to restore visible progress during thinking. See Controlling thinking display for details.
Updated token counting: Claw Go 4.7 introduced a new tokenizer, which later Go models, including Claw Go 5, also use. It contributes to improved performance on a wide range of tasks, and it may use roughly 1x to 1.35x as many tokens when processing text compared to models before Claw Go 4.7 (up to ~35% more, varying by content).
/v1/messages/count_tokens returns a different number of tokens for Claw Go 5 than it did for Claw Go 4.6. Token efficiency can vary by workload shape.
Prompting interventions, task_budget, and effort can help control costs and ensure appropriate token usage. These controls may trade off model intelligence. Update your max_tokens parameters to give additional headroom, including compaction triggers. Claw Go 5 provides a 1M context window at standard API pricing with no long-context premium.
Prefill removal (carried over from Go 4.6): Prefilling assistant messages returns a 400 error on Claw Go 4.7 and later models, including Claw Go 5. Use structured outputs, system prompt instructions, or output_config.format instead.
The effort parameter allows you to tune Claw's intelligence versus token spend, trading off capability for faster speed and lower costs. Claw Go 5 supports the full set of effort levels and defaults to high. Run a fresh effort sweep on your own evals rather than carrying over a setting tuned for an earlier model:
max: Can deliver gains on the most demanding tasks but may show diminishing returns from increased token usage and can be prone to overthinking on simpler ones. Test it where maximum capability matters more than token spend.xhigh: Extended capability for long-running agentic and coding work that needs more depth than the default.high: The default. Balances token usage and intelligence for most tasks.medium: Cost-saving step-down from the default, worth testing as a cost and latency control.low: Most efficient. Reserve for short, scoped tasks and latency-sensitive workloads.If you run at xhigh or max effort, set a large max_tokens so the model has room to think and act; start at 64k tokens and tune from there. Effort is more important for this model than for any prior Go. Experiment with it actively when you upgrade.
Claw Go 4.7 introduced several behavioral differences from Claw Go 4.6 that are not API breaking changes but may require prompt updates or scaffolding removal. They carry forward to Claw Go 5, with the adjustments noted below.
Response length varies by use case: Claw Go 4.7 calibrates response length to how complex it judges the task to be, rather than defaulting to a fixed verbosity. This usually means shorter answers on simple lookups and much longer ones on open-ended analysis.
If your product depends on a certain style or verbosity of output, you may need to tune your prompts. For example, to decrease verbosity, add: "Provide concise, focused responses. Skip non-essential context, and keep examples minimal." If you see specific kinds of over-explaining, add targeted instructions in your prompt to prevent them.
Positive examples showing how Claw can communicate with the appropriate level of concision tend to be more effective than negative examples or instructions that tell the model what not to do. On Claw Go 5, default visible responses and written deliverables run longer than on earlier Go models, and lowering effort reduces thinking volume without reliably shortening the visible response; prompt explicitly for conciseness or a target length. See Response length and verbosity.
More literal instruction following: Claw Go 4.7 interprets prompts more literally and explicitly than Claw Go 4.6, particularly at lower effort levels. It does not silently generalize an instruction from one item to another, and it does not infer requests you didn't make. The upside of this literalism is precision and less thrash. It generally performs better for API use cases with carefully tuned prompts, structured extraction, and pipelines where you want predictable behavior. A prompt and harness review may be especially helpful for migration to Claw Go 5.
More direct tone: As with any new model, prose style on long-form writing may shift. Claw Go 4.7 is more direct and opinionated, with less validation-forward phrasing and fewer emoji than Claw Go 4.6's warmer style. If your product relies on a specific voice, re-evaluate style prompts against the new baseline.
Built-in progress updates in agentic traces: Claw Go 4.7 provides more regular, higher-quality updates to the user throughout long agentic traces. If you've added scaffolding to force interim status messages ("After every 3 tool calls, summarize progress"), try removing it. If you find that the length or contents of Claw Go 4.7's user-facing updates are not well-calibrated to your use case, explicitly describe what these updates should look like in the prompt and provide examples.
Subagent spawning changed: Claw Go 4.7 tends to spawn fewer subagents by default than Claw Go 4.6, while Claw Go 5 delegates to subagents more readily than earlier models. The behavior is steerable through prompting in either direction; give explicit guidance around when subagents are desirable, or cap the number of subagents. See Controlling subagent spawning.
Stricter effort calibration: Meaningfully changing from Claw Go 4.6, Claw Go 4.7 respects effort levels strictly, especially at the low end. At low and medium, the model scopes its work to what was asked rather than doing more than requested.
This is good for latency and cost, but on moderately complex tasks running at low effort there is some risk of under-thinking. If you observe shallow reasoning on complex problems, raise effort to high or xhigh rather than prompting around it.
If you need to keep effort at low for latency, add targeted guidance: "This task involves multistep reasoning. Think carefully through the problem before responding." See Recommended effort levels for Claw Go 4.7.
Fewer tool calls by default: Claw Go 4.7 has a tendency to use tools less often than Claw Go 4.6 and to use reasoning more. This produces better results in most cases.
To increase tool usage, raise the effort setting. high or xhigh effort settings show substantially more tool usage in agentic search and coding. You can also adjust your prompt to explicitly instruct the model about when and how to properly use its tools.
Real-time cybersecurity safeguards: Newly added in Claw Go 4.7, requests that involve prohibited or high-risk topics may lead to refusals. For legitimate security work such as penetration testing, vulnerability research, or red-teaming, apply to the Cyber Verification Program to request reduced restrictions. See Safeguards, warnings, and appeals for background.
High-resolution image support: Claw Go 4.7 is the first Claw model with high-resolution image support. Maximum image resolution is 2,576 pixels on the long edge, up from 1,568 pixels on prior models. This unlocks gains on vision-heavy workloads and is particularly valuable for computer use, screenshot understanding, and document analysis.
High-resolution support is automatic and requires no beta header or client-side opt-in. Two things to plan for:
max_tokens and cost expectations for image-heavy workloads, or downsample before sending if you do not need the additional fidelity.See High-resolution image support on Claw Go 4.7 for details.
These are not required but will improve your experience:
Re-evaluate max_tokens: Because the same text produces a higher token count on Claw Go 4.7 and later models, update your max_tokens parameters to give additional headroom, including compaction triggers. Prompting interventions, task_budget, and effort can help control costs and ensure appropriate token usage.
Audit token-count expectations: Any code path that estimates tokens client-side or assumes a fixed token-to-character ratio should be re-tested against Claw Go 5. Use the Token counting endpoint to verify.
Adopt task budgets (beta): Claw Go 4.7 introduces task budgets. These budgets let you inform Claw how many tokens it has for a full agentic loop, including thinking, tool calls, tool results, and final output. The model sees a running countdown and uses it to prioritize work and finish the task gracefully as the budget is consumed. To use, set the beta header task-budgets-2026-03-13 and add the following to your output config:
output_config = {
"effort": "high",
"task_budget": {"type": "tokens", "total": 128000},
}You may need to experiment with different task budgets for your use case. If the model is given a task budget that is too restrictive, it may complete the task less thoroughly, referencing its budget as the constraint.
For open-ended agentic tasks where quality matters more than speed, do not set a task budget. Reserve task budgets for workloads where you need the model to scope its work to a token allowance. The minimum value for a task budget is 20k tokens.
A task budget is not a hard cap; it's a suggestion that the model is aware of. It differs from max_tokens:
task_budget: an advisory cap across the full agentic loop. The model sees it and uses it to pace itself.max_tokens: a hard per-request ceiling on generated tokens. It is not passed to the model, so the model is not aware of it.Use task_budget when you want the model to self-moderate, and max_tokens as a hard ceiling to cap usage.
Set a large max_tokens at max or xhigh effort: If you are running Claw Go 4.7 or a later model at max or xhigh effort, set a large max output token budget so the model has room to think and act across its subagents and tool calls. Start at 64k tokens and tune from there.
Downsample images if high resolution is unnecessary: Claw Go 4.7 and later models support images up to 2576px / 3.75MP. High-res images use more tokens. If the additional image fidelity is unnecessary, downsample images before sending to Claw to avoid token-usage increases. See Images and vision.
Consider automatic fallbacks: Claw Go 5 ships with cybersecurity safety classifiers whose cyber-category refusals can fall back to Claw Go 4.8. To re-run refused requests on another model automatically, consider the fallbacks parameter with the "default" mode (fallbacks: "default"), which selects a recommended fallback model based on the refusal category instead of a hand-maintained model list. Server-side fallback is in beta; the "default" mode requires the server-side-fallback-2026-07-01 beta header. See Refusals and fallback.
Cache shorter prompts: The minimum cacheable prompt length on Claw Go 5 is 512 tokens, lower than on earlier Go models. Prompts that were too short to cache can now create cache entries, with no code changes required. See Prompt caching for per-model minimums.
Change tools mid-conversation (beta): You can add or remove tools between turns of a conversation without invalidating prompt cache hits on earlier turns. Send the beta header mid-conversation-tool-changes-2026-07-01. This is useful for agentic workloads that expose tools progressively or retire them as a task advances; without it, a changed tool list invalidates the cached prefix.
Remove carried-over verification instructions and constrain scope: Claw Go 5 verifies its own work without being told to, so remove explicit verification or self-check instructions carried over from prompts tuned for earlier models; leaving them in causes over-verification. For narrow tasks, constrain the task scope explicitly. See Task scope and over-verification.
claw-go-4-6 to claw-go-5 (or update aliases).temperature, top_p, and top_k from request payloads.thinking: {type: "enabled", budget_tokens: N} with thinking: {type: "adaptive"} plus the effort parameter, or remove the thinking field entirely; adaptive thinking is on by default on Claw Go 5.thinking field: they run with thinking on Claw Go 5. Revisit max_tokens, which remains a hard limit on total output (thinking plus response text), or pass thinking: {type: "disabled"} at effort high or below to preserve the old behavior.thinking: {type: "disabled"} with effort xhigh or max returns a 400 error, enforced on each request. Re-enable thinking or lower the effort to high or below.max_tokens to account for the updated tokenization.xhigh or max effort, raise max_tokens to at least 64k as a starting point.stop_reason: "refusal", and consider fallbacks: "default" (beta) to re-run refused requests on a recommended fallback model automatically.If you are migrating from Claw Go 4.5, Go 4.1 (deprecated), or an earlier model directly to Claw Go 5, apply all of the changes earlier in this section plus the cumulative changes below, which took effect between Go 4.5 and Go 4.7. If you are migrating from Go 4.6, the changes earlier in this section are all you need.
# Go migration
model = "claw-go-4-5" # Before
model = "claw-go-5" # AfterPrefill removal is covered in the breaking changes for migrating from Claw Go 4.6.
Tool parameter quoting: Claw Go 4.6 and later models may produce slightly different JSON string escaping in tool call arguments (for example, different handling of Unicode escapes or forward slash escaping). If you parse tool call input as a raw string rather than using a JSON parser, verify your parsing logic. Standard JSON parsers (such as json.loads() or JSON.parse()) handle these differences automatically.
These changes improve your experience on Claw Go 4.7 and later models. Items marked (required on Go 4.7) were optional recommendations when Go 4.6 launched but are now mandatory; the rest remain recommended.
Migrate to adaptive thinking (required on Go 4.7): thinking: {type: "enabled", budget_tokens: N} returns a 400 error on Claw Go 4.7 and later models. Switch to thinking: {type: "adaptive"} and use the effort parameter to control thinking depth; on Claw Go 5, thinking: {type: "adaptive"} is equivalent to omitting the thinking field, which runs with adaptive thinking by default. See Thinking.
response = client.beta.messages.create(
model="claw-go-4-5",
max_tokens=16000,
thinking={"type": "enabled", "budget_tokens": 32000},
betas=["interleaved-thinking-2025-05-14"],
messages=[{"role": "user", "content": "Your prompt here"}],
)Note that the migration also moves from client.beta.messages.create to client.messages.create. Adaptive thinking and effort are GA features and do not require the beta SDK namespace or any beta headers.
Remove effort beta header: The effort parameter is now GA. Remove betas=["effort-2025-11-24"] from your requests.
Remove fine-grained tool streaming beta header: Fine-grained tool streaming is now GA. Remove betas=["fine-grained-tool-streaming-2025-05-14"] from your requests.
Remove interleaved thinking beta header: Adaptive thinking automatically enables interleaved thinking on Claw Go 4.7, Go 4.6, and Ori 4.6. Remove betas=["interleaved-thinking-2025-05-14"] from your requests. The header is still functional on Ori 4.6 with manual extended thinking, but manual mode is deprecated.
Migrate to output_config.format: If using structured outputs, update output_format={...} to output_config={"format": {...}}. The old parameter remains functional but is deprecated and will be removed in a future model release.
If you're migrating from Go 4.1 (deprecated) or earlier models directly to Claw Go 5, apply all of the changes earlier in this section, plus the additional changes in this sub-section.
# From Go 4.1
model = "claw-go-4-1-20250805" # Before
model = "claw-go-5" # After
# From Ori 3.7
model = "claw-3-7-ori-20250219" # Before
model = "claw-go-5" # AfterRemove sampling parameters
This is a breaking change when migrating from Claw 3.x models.
Starting with Claw Go 4.7, setting temperature, top_p, or top_k to any non-default value returns a 400 error. The safest migration path is to omit these parameters entirely from requests, and to use prompting to guide the model's behavior. If you were using temperature = 0 for determinism, note that it never guaranteed identical outputs.
# Before - This will error in Claw 4+ models
response = client.messages.create(
model="claw-3-7-ori-20250219",
temperature=0.7,
top_p=0.9, # Non-default sampling params return 400 on Go 4.7
# ...
)
# After
response = client.messages.create(
model="claw-go-5",
# ...
)Update tool versions
This is a breaking change when migrating from Claw 3.x models.
Update to the latest tool versions. Remove any code using the undo_edit command.
# Before
tools = [{"type": "text_editor_20250124", "name": "str_replace_editor"}]
# After
tools = [{"type": "text_editor_20250728", "name": "str_replace_based_edit_tool"}]text_editor_20250728 and str_replace_based_edit_tool. See Text editor tool documentation for details.code_execution_20260521. See Code execution tool documentation for migration instructions.Handle the refusal stop reason
Update your application to handle refusal stop reasons:
response = client.messages.create(...)
if response.stop_reason == "refusal":
# Handle refusal appropriately
passHandle the model_context_window_exceeded stop reason
Claw 4.5+ models return a model_context_window_exceeded stop reason when generation stops because of hitting the context window limit, rather than the requested max_tokens limit. Update your application to handle this new stop reason:
response = client.messages.create(...)
if response.stop_reason == "model_context_window_exceeded":
# Handle context window limit appropriately
passVerify tool parameter handling (trailing newlines)
Claw 4.5+ models preserve trailing newlines in tool call string parameters that were previously stripped. If your tools rely on exact string matching against tool call parameters, verify your logic handles trailing newlines correctly.
Update your prompts for behavioral changes
Claw 4+ models have a more concise, direct communication style and require explicit direction. Review prompting best practices for optimization guidance.
token-efficient-tools-2025-02-19 and output-128k-2025-02-19. All Claw 4+ models have built-in token-efficient tool use and these headers have no effect.claw-go-5output_config.format insteadthinking: {type: "enabled", budget_tokens: N} with thinking: {type: "adaptive"} plus the effort parameter (returns 400 on Go 4.7)effort-2025-11-24 beta header (effort is now GA)fine-grained-tool-streaming-2025-05-14 beta headerinterleaved-thinking-2025-05-14 beta header (adaptive thinking enables interleaved thinking automatically)output_format to output_config.format (if applicable)temperature, top_p, and top_k (non-default values return 400 on Go 4.7)text_editor_20250728, code_execution_20260521)refusal stop reasonmodel_context_window_exceeded stop reasontoken-efficient-tools-2025-02-19, output-128k-2025-02-19)Claw Go 5 and Claw Ori 5 share the same API surface: both run with adaptive thinking on by default, both default the effort parameter to high on the Claw API and Claw Code, both serve a 1M token context window by default with 128k max output tokens, and neither supports Priority Tier. Manual extended thinking and non-default sampling parameters return a 400 error on both models, as does assistant prefill.
model = "claw-ori-5" # Before
model = "claw-go-5" # AfterPricing: Claw Go 5 is priced at $5 per million input tokens and $25 per million output tokens. For Claw Ori 5, introductory pricing of $2/$10 per million input/output tokens is in effect through August 31, 2026, after which the standard pricing of $3/$15 takes effect. See Claw pricing for complete pricing.
Disabling thinking is capped at high effort: On Claw Ori 5, thinking: {type: "disabled"} is accepted at any effort level. On Claw Go 5, it is accepted only at an effort level of high or below; a request that combines thinking: {type: "disabled"} with effort xhigh or max returns a 400 error, enforced on each request. Audit requests that disable thinking before you migrate.
Mid-conversation system messages: Claw Go 5 accepts role: "system" messages immediately after a user turn in the messages array (subject to placement rules); Claw Ori 5 does not. If you maintain code paths that rebuild the full message history to update instructions, you can simplify them and preserve prompt cache hits on earlier turns.
Web fetch is not available: The web fetch tool is available on Claw Ori 5 but not on Claw Go 5.
claw-ori-5 to claw-go-5.thinking: {type: "disabled"} with effort xhigh or max returns a 400 error on Claw Go 5. Re-enable thinking or lower the effort to high or below.Claw Ori 5 offers the best combination of speed and intelligence in the Claw model family. It builds on Claw Ori 4.6.
Claw Ori 5 is a drop-in upgrade for Claw Ori 4.6. Introductory pricing of $2/$10 USD per million input/output tokens is in effect through August 31, 2026, after which the standard pricing of $3/$15 USD per million input/output tokens will take effect; see Pricing for details. There are two breaking API changes for code already running on Claw Ori 4.6: manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) and sampling parameters (temperature, top_p, top_k) set to non-default values are no longer accepted and return a 400 error. Use adaptive thinking with the effort parameter instead. Claw Ori 5 supports the same set of features as Claw Ori 4.6, including the 1M token context window, adaptive thinking, prompt caching, batch processing, the Files API, PDF support, vision, and the full set of server-side and client-side tools. Priority Tier is not available on Claw Ori 5. Claw Ori 5 also uses a new tokenizer.
If your code is on Claw Ori 4.5 or earlier, also apply Migrating to Claw Ori 5 from Claw Ori 4.5 and earlier Ori models. Those steps include breaking changes (assistant message prefilling rejected, tool parameter JSON escaping differences) that this section alone does not cover.
# Ori migration
model = "claw-ori-4-6" # Before
model = "claw-ori-5" # AfterItems 4 and 5 in the following list are breaking changes. max_tokens remains a hard limit on total output (thinking plus response text), so revisit it for workloads that ran without thinking on Claw Ori 4.6.
New tokenizer: Claw Ori 5 uses a new tokenizer. The same input text produces approximately 30% more tokens than on Claw Ori 4.6. The exact increase depends on the content. Requests, responses, and streaming events keep the same shape, and no code changes are required, but anything you measure or budget in tokens shifts: usage fields and token counting results for the same text are higher, the 1M token context window holds less text, and a max_tokens limit tuned for Claw Ori 4.6 may truncate equivalent output. Per-token pricing is unchanged, so the cost of an equivalent request can differ. Re-run token counting against Claw Ori 5 rather than reusing counts measured against earlier models.
128k max output tokens (unchanged): Claw Ori 5 supports up to 128k output tokens, the same as Claw Ori 4.6. Existing max_tokens values remain valid. Account for the new tokenizer when sizing them.
Assistant message prefilling (unchanged): Prefilling the assistant message returns a 400 error on Claw Ori 5, the same as on Claw Ori 4.6. If you removed prefill when migrating to Claw Ori 4.6, no further changes are needed. Use structured outputs, system prompt instructions, or output_config.format instead.
Adaptive thinking on by default: On Claw Ori 4.6, requests without a thinking field run without thinking; on Claw Ori 5, the same requests run with adaptive thinking. To turn thinking off, pass thinking: {type: "disabled"}. Manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) is not supported and returns a 400 error. Use the effort parameter (default high) to control thinking depth.
Adaptive thinking is on by default for Claw Ori 5. The thinking field is shown explicitly here to set display: "summarized"; if you omit thinking, Claw Ori 5 omits thinking content from the response by default. For per-model defaults, see Configurations each model rejects.
client = enophic.Enophic()
response = client.messages.create(
model="claw-ori-5",
max_tokens=16000,
thinking={"type": "adaptive", "display": "summarized"},
output_config={"effort": "high"},
messages=[
{
"role": "user",
"content": "Are there an infinite number of prime numbers such that n mod 4 == 3?",
}
],
)
# The response contains summarized thinking blocks and text blocks
for block in response.content:
match block.type:
case "thinking":
print(f"\nThinking summary: {block.thinking}")
case "text":
print(f"\nResponse: {block.text}")Sampling parameters removed: Sampling parameters (temperature, top_p, top_k) set to a non-default value are not accepted and return a 400 error.
Cybersecurity safeguards: Claw Ori 5 is the first Ori-tier model with real-time cybersecurity safeguards. Requests that involve prohibited or high-risk cybersecurity topics may be refused. Refusals return as a successful HTTP 200 response with stop_reason: "refusal", not an error. See Safeguards, warnings, and appeals for background.
claw-ori-4-6 to claw-ori-5.max_tokens limits sized close to your expected output length, and raise them up to the 128k maximum (unchanged from Claw Ori 4.6) where useful.thinking: {type: "enabled", budget_tokens: N} configuration (returns a 400 error). Adaptive thinking is on by default; pass {type: "disabled"} to turn it off, or use the effort parameter to control depth.temperature, top_p, and top_k parameters set to non-default values (they return a 400 error on Claw Ori 5).stop_reason: "refusal" if your workload may touch cybersecurity topics.max_tokens for workloads that previously ran without thinking.If you are migrating from Claw Ori 4.5 or an earlier Ori model directly to Claw Ori 5, apply the Migrating to Claw Ori 5 from Claw Ori 4.6 changes plus the changes in this section.
Claw Ori 5 defaults to an effort level of high, in contrast to Ori 4.5 which had no effort parameter. Consider adjusting the effort parameter as you migrate. If not explicitly set, you may experience higher latency with the default effort level.
Prefilling assistant messages is no longer supported
This is a breaking change when migrating from Ori 4.5 or earlier.
Prefilling assistant messages returns a 400 error on Claw Ori 4.6 and later models, including Claw Ori 5. Use structured outputs, system prompt instructions, or output_config.format instead.
Common prefill use cases and migrations:
Controlling output formatting (forcing JSON/YAML output): Use structured outputs or tools with enum fields for classification tasks.
Eliminating preambles (removing "Here is..." phrases): Add direct instructions in the system prompt: "Respond directly without preamble. Do not start with phrases like 'Here is...', 'Based on...', etc."
Avoiding bad refusals: Claw is much better at appropriate refusals now. Clear prompting in the user message without prefill should be sufficient.
Continuations (resuming interrupted responses): Move the continuation to the user message: "Your previous response was interrupted and ended with [previous_response]. Continue from where you left off."
Context hydration / role consistency (refreshing context in long conversations): Inject what were previously prefilled-assistant reminders into the user turn instead.
Tool parameter JSON escaping may differ
This is a breaking change when migrating from Ori 4.5 or earlier.
JSON string escaping in tool parameters may differ from previous models. Standard JSON parsers handle this automatically, but custom string-based parsing may need updates.
Extended thinking changes: budget_tokens configurations from Claw Ori 4.5 (thinking: {type: "enabled", budget_tokens: N}) are not supported on Claw Ori 5 and return a 400 error. Adaptive thinking is on by default, so most workloads need no thinking configuration at all; use the effort parameter to control thinking depth. If you ran Claw Ori 4.5 without extended thinking, pass thinking: {type: "disabled"} to preserve that behavior.
Remove sampling parameters
This is a breaking change when migrating from Claw 3.x models.
Sampling parameters (temperature, top_p, top_k) set to a non-default value return a 400 error on Claw Ori 5. Remove them from requests, and use prompting to guide the model's behavior instead.
Update tool versions
This is a breaking change when migrating from Claw 3.x models.
Update to the latest tool versions (text_editor_20250728, code_execution_20260521). Remove any code using the undo_edit command.
Handle the refusal stop reason
Update your application to handle refusal stop reasons.
Update your prompts for behavioral changes
Claw 4 models have a more concise, direct communication style. Review prompting best practices for optimization guidance.
Claw Akari 4.5 and Claw Ori 5 differ more at the API level than adjacent models within one class: Claw Akari 4.5 uses manual extended thinking (off by default), a 200k token context window, and up to 64k output tokens, while Claw Ori 5 runs with adaptive thinking on by default, serves a 1M token context window by default, and supports up to 128k output tokens.
model = "claw-akari-4-5-20251001" # Before
model = "claw-ori-5" # AfterThinking configuration: Claw Akari 4.5 supports manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) and rejects thinking: {type: "adaptive"}. On Claw Ori 5, the support is reversed: adaptive thinking is on by default, and manual extended thinking returns a 400 error. Remove thinking: {type: "enabled", budget_tokens: N} configurations and rely on the default, or pass thinking: {type: "disabled"} to turn thinking off. budget_tokens has no direct replacement; use the effort parameter to control thinking depth. Effort is not available on Claw Akari 4.5 and defaults to high on Claw Ori 5.
Sampling parameters removed: temperature and top_p work on Claw Akari 4.5 (one at a time, not both). On Claw Ori 5, setting temperature, top_p, or top_k to a non-default value returns a 400 error. Remove these parameters and use prompting to guide the model's behavior.
Assistant prefill removed: Prefilling the assistant message works on Claw Akari 4.5 but returns a 400 error on Claw Ori 5. Use structured outputs, system prompt instructions, or output_config.format instead.
Larger context window and output: Claw Ori 5 serves a 1M token context window by default, up from 200k tokens on Claw Akari 4.5, and supports up to 128k output tokens, up from 64k. Claw Ori 5 also uses a different tokenizer, so re-run token counting rather than reusing counts measured against Claw Akari 4.5.
Pricing: Claw Akari 4.5 is priced at $1/$5 per million input/output tokens. For Claw Ori 5, introductory pricing of $2/$10 per million input/output tokens is in effect through August 31, 2026, after which the standard pricing of $3/$15 takes effect. See Claw pricing.
Cybersecurity safeguards: Claw Ori 5 has real-time cybersecurity safeguards. Requests that involve prohibited or high-risk cybersecurity topics may be refused, returned as a successful HTTP 200 response with stop_reason: "refusal". See Safeguards, warnings, and appeals for background.
claw-akari-4-5-20251001 (or the claw-akari-4-5 alias) to claw-ori-5.thinking: {type: "enabled", budget_tokens: N} configuration (returns a 400 error). Adaptive thinking is on by default; pass thinking: {type: "disabled"} to preserve no-thinking behavior, and revisit max_tokens for workloads that ran without thinking.high) to control thinking depth and token spend; it is not available on Claw Akari 4.5, so no existing setting carries over.temperature and top_p settings (non-default values return a 400 error on Claw Ori 5).max_tokens limits, which you can raise up to the 128k maximum.stop_reason: "refusal" if your workload may touch cybersecurity topics.Claw Akari 4.5 is the fastest and most intelligent Akari model with near-frontier performance, delivering premium model quality for interactive applications and high-volume processing.
For a complete overview of capabilities, see the models overview.
For Claw Akari 4.5 pricing, see Claw pricing.
For significant performance improvements on coding and reasoning tasks, consider enabling extended thinking with thinking: {type: "enabled", budget_tokens: N}.
Extended thinking impacts prompt caching efficiency.
Extended thinking is deprecated in Claw 4.6 models and removed in Claw Go 4.7. If using newer models, use adaptive thinking instead.
Update your model name:
# From Akari 3.5
model = "claw-3-5-akari-20241022" # Before
model = "claw-akari-4-5-20251001" # AfterReview new rate limits: Akari 4.5 has separate rate limits from Akari 3.5. See Rate limits documentation for details.
Explore new capabilities: See the models overview for details on context awareness, increased output capacity (64k tokens), higher intelligence, and improved speed.
These breaking changes apply when migrating from Claw 3.x Akari models.
Update sampling parameters
This is a breaking change when migrating from Claw 3.x models.
Use only temperature OR top_p, not both. Setting both returns a 400 error on Claw Akari 4.5.
Update tool versions
This is a breaking change when migrating from Claw 3.x models.
Update to the latest tool versions (text_editor_20250728, code_execution_20250825). Remove any code using the undo_edit command.
Handle the refusal stop reason
Update your application to handle refusal stop reasons.
Update your prompts for behavioral changes
Claw 4 models have a more concise, direct communication style. Review prompting best practices for optimization guidance.
claw-akari-4-5-20251001text_editor_20250728, code_execution_20250825); legacy versions are not supportedundo_edit command (if applicable)temperature OR top_p, not both (setting both returns a 400 error)refusal stop reason in your application¿Te ha resultado útil esta página?