Claude Mythos 5.1 vs Fable 5.1: Same Model, Different Safeguards

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 together on September 1, 2026, describing them as “the same model but with different levels of safeguards.” They use the same weights, specifications, pricing, and API. Fable 5.1 is generally available, while Mythos 5.1 is limited to approved organizations in Project Glasswing through cybersecurity and life-sciences verification programs. Mythos 5.1 also powers Anthropic’s Claude Security product.

Try Apidog today

This guide compares the models, explains their four practical differences, shows how Mythos 5.1 access works, and examines the only published benchmark gap. Sources include Anthropic’s launch post and the Fable 5.1 migration guide, which covers both models.

For more context, see what “Mythos-class” means and Fable 5 vs. Mythos 5.

Side-by-side comparison

	Claude Fable 5.1	Claude Mythos 5.1
Model ID	claude-fable-5-1	claude-mythos-5-1
Availability	All customers	Project Glasswing participants only
Access route	Sign up	Cyber Verification Program or Life Sciences Verification Program; US organizations
Pricing	$10 / $50; cache reads $0.25	Same
Context / maximum output	1M / 128K	Same
Knowledge cutoff	June 2026	Same
Thinking	Adaptive, always on	Same
Forced tool_choice	400	400
Safety classifiers	Full Fable set: cyber, bio, frontier_llm, reasoning_extraction, general_harms	Program-specific safeguards; more permissive for approved defensive work
Vulnerability work	Can identify vulnerabilities but cannot develop exploits	Can discover exploit paths for approved defensive research
Reads Fable 5.1 thinking blocks	Yes	Yes—the only other model that can
History-editing check	Yes	No
Platforms	Claude API, Bedrock, Claude Platform on AWS, Google Cloud, Foundry	Claude API, Bedrock, Google Cloud, Foundry; not Claude Platform on AWS
Priority Tier	No	No
Data retention	30-day required; Covered Model	Same
Terminal-Bench 4.0	55.8%	60.9%

What is identical

Anthropic describes Fable 5.1 and Mythos 5.1 as the same model with different safeguards. Both provide:

• A 1M-token context window
• A 128K-token maximum output
• Always-on adaptive thinking
• Five effort levels
• A June 2026 knowledge cutoff
• $0.25 cache reads
• The same pricing and API behavior
• The same three Fable 5-generation breaking changes, except for the history-editing check described below

The Fable 5.1 specification overview also applies to Mythos 5.1.

Both models require 30-day data retention and are Covered Models. They are not available under zero data retention unless Anthropic explicitly authorizes it. Neither supports Priority Tier, and both include Anthropic’s statistical text watermark on every platform.

Difference 1: availability and access

Fable 5.1 is generally available through the Claude API and every supported partner platform.

Mythos 5.1 is limited to approved Project Glasswing customers. At launch, there were two access routes:

1. The Cyber Verification Program for defensive security professionals, with applications submitted through Anthropic’s portal.
2. The Life Sciences Verification Program for life-sciences organizations, operated in partnership with the US government with enrollment expanding over time.

Both programs were limited to US organizations at launch. Access is arranged through your Anthropic, AWS, or Google Cloud account team. Whether existing Mythos 5 access transfers automatically was still being confirmed.

This differs from the Mythos 5 generation, which was invitation-only and ran no safety classifiers. Mythos 5.1 still has safeguards; they are calibrated to the access program rather than removed.

Difference 2: safeguard behavior

Fable 5.1 runs the complete Fable classifier set. When it declines a request, the API returns HTTP 200 with:

{
  "stop_reason": "refusal"
}

The response also includes a refusal category. Anthropic reports substantial reductions in false positives compared with Fable 5:

• Biology classifiers trigger 85% less often on benign requests.
• Cyber safeguards produce approximately 60% fewer interventions per Claude Code session.

Fable 5.1 can identify software vulnerabilities, but it will not develop exploits.

Mythos 5.1 is intended for work that Fable 5.1 may decline, including exploit-path discovery for approved defensive research and life-sciences work that triggers the bio classifier. It still runs safeguards, so your integration should handle refusals on both models:

if (response.stop_reason === "refusal") {
  const category = response.stop_details?.category;
  // Log, retry, or route to an approved fallback.
}

Anthropic’s launch post and system card also describe:

• A real-time classifier for aggressive probing and sandbox-escape attempts
• Stronger isolation for high-risk cyber workloads
• New network-isolation requirements for external evaluators

These measures follow incidents in earlier evaluations where models reached real systems.

Difference 3: history-editing validation

Fable 5.1 thinking blocks are valid only in the exact conversation that produced them. On accounts created on or after August 31, 2026, editing any of the following invalidates subsequent thinking blocks:

• An earlier conversation turn
• The system prompt
• The tools array

The API then returns a 400 error because the blocks are bound to a different conversation. The preserved thinking guide covers this behavior in detail.

Mythos 5.1 does not run this history-editing check. However, thinking blocks remain bound to their producing model:

• Mythos 5.1 can read Mythos 5 thinking blocks.
• Mythos 5 cannot read Mythos 5.1 thinking blocks.
• Editing history still restarts the prompt cache.
• History edits do not produce the “bound to a different conversation” 400 error.

Mythos 5.1 is also the only model besides Fable 5.1 that can read Fable 5.1 thinking blocks. You can move a conversation between those two models without losing its reasoning. Switching to Opus 5 causes the API to drop the blocks.

Difference 4: platforms and rate limits

Fable 5.1 is available on Claude Platform on AWS. Mythos 5.1 is not.

Mythos 5.1 is available through:

• The Claude API
• Amazon Bedrock as anthropic.claude-mythos-5-1
• Google Cloud
• Microsoft Foundry

On Bedrock, Mythos 5.1 is available in us-east-1 but is not publicly listed.

The models also use separate rate-limit pools:

• Fable 5.1 shares the Fable 5.x pool with Fable 5.
• Mythos 5.1 shares the Mythos pool with Mythos 5.

What the benchmark gap shows

Anthropic published one benchmark result for Mythos 5.1:

• Mythos 5.1: 60.9% on Terminal-Bench 4.0
• Fable 5.1: 55.8%

Because the underlying model is the same, the five-point difference is the clearest public measurement of safeguard overhead on an agentic coding task unrelated to cyber or biology. The classifiers presumably intervene on some benign steps during long terminal sessions.

In other words, Mythos 5.1 is a demonstrably stronger option for some workloads, even though it is not a different model. The complete benchmarks breakdown covers the rest of the results.

Anthropic’s science claims were all made for Mythos 5.1, not Fable 5.1:

• Protein binders with affinities 10 times higher than the best competition submissions across three targets
• A nearly 50% hit rate across 12 targets, compared with a typical 10–15%
• A Venus elevation map with detail down to two or three kilometers

These are the workloads targeted by the Life Sciences Verification Program.

Which model should you use?

For most developers, the answer is Fable 5.1 because it is generally available.

Mythos 5.1 is relevant if your organization:

• Performs defensive security research that triggers the cyber classifier
• Conducts legitimate life-sciences research that triggers the bio classifier
• Has been approved for Project Glasswing

If you qualify, migrating from Fable 5.1 should usually require only a model-ID change:

{
  "model": "claude-mythos-5-1"
}

The API behavior is otherwise the same, apart from history-editing validation.

Before applying for Mythos access, try these steps if Fable 5.1 is refusing legitimate defensive work:

1. Follow Anthropic’s phrasing guidance. Ask “are there any bugs?” instead of “does this compile?”
2. Include documentation for lesser-known programming languages.
3. Remove tools that inject base64 data into the context.
4. Configure a fallback:

{
  "fallbacks": "default"
}

This allows declined requests to fall through to Opus 5 or Opus 4.8. See the refusal handling guide for implementation details.

Test both model IDs

Because the models are otherwise identical, the easiest comparison is to send the same request body to both IDs in Apidog.

First, send a benign vulnerability-analysis prompt to:

• claude-fable-5-1
• claude-mythos-5-1, if your organization has access

Compare:

• stop_reason
• stop_details.category

Then send a two-request sequence that edits the system prompt between turns while including the thinking-binding-controls-2026-08-01 header.

Expected behavior:

• Fable 5.1 reports prefix_binding_mismatch in input_transformations.
• Mythos 5.1 does not run that history-editing check.

Download Apidog to run the comparison.

FAQ

Is Claude Mythos 5.1 more capable than Fable 5.1?

They use the same underlying model, so their capabilities are the same. The published Terminal-Bench 4.0 results—60.9% for Mythos 5.1 versus 55.8% for Fable 5.1—reflect safeguard interventions rather than different model weights.

How do I get access to Claude Mythos 5.1?

Apply through Project Glasswing’s Cyber Verification Program or Life Sciences Verification Program. Both programs were limited to US organizations at launch. You can also contact your Anthropic, AWS, or Google Cloud account team.

Does Mythos 5.1 cost more than Fable 5.1?

No. Both cost $10 and $50 per million tokens, with the same cache and batch rates.

Does Claude Mythos 5.1 refuse requests?

Yes. Unlike Mythos 5, which ran no classifiers, Mythos 5.1 uses safeguards calibrated to the access program. Handle stop_reason: "refusal" and stop_details.category for both models.

Can I switch a conversation between Fable 5.1 and Mythos 5.1?

Yes. They are the only two models that can read each other’s thinking blocks, so switching between them does not discard the reasoning.

Is Mythos 5.1 available on Amazon Bedrock?

Yes. It is available as anthropic.claude-mythos-5-1 in us-east-1, although it is not publicly listed. It is not available on Claude Platform on AWS.