Released September 1, 2026, Mythos 5.1 matches Fable 5.1 on context (1 million tokens), output (128,000 tokens) and price, but relaxes cyber and biology safeguards for verified users. Anthropic's October 2026 update gave its Cyber Verification Program Defense, Red Team and Specialized tiers. It suits vetted security and life-science teams, not general API customers.
What changed
Claude Mythos 5.1 is Anthropic's verification-gated model, identical to Claude Fable 5.1 apart from safeguards. ARC Prize measured those shared weights at 90.0% on ARC-AGI-2 at maximum effort. Only organizations verified through Anthropic's programs, such as the Cyber Verification Program, can use it, with pricing identical to Fable 5.1.
Where it sits
- $20.00/M$ per 1M tokensBlended price (3:1)Lower is better#74 / 79peer median $1.71/Mvendor price, checked by HokAI
Pricier than 97% of the 79 GA models with a published price, and one of 31 that document a zero-data-retention option.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: Anthropic · Family: Claude Mythos
Context window: 1,000,000 tokens · Max output: 128,000
Input modalities: text, image · Output: text
About Claude Mythos 5.1
Anthropic released Claude Mythos 5.1 on September 1, 2026 as a restricted-access large language model, one of the frontier entries in the AI models directory. Anthropic's docs give it the same specifications and pricing as Claude Fable 5.1; its announcement calls the two identical apart from safeguards: Fable 5.1 is generally available with standard cybersecurity and biology classifiers, while Mythos 5.1 offers more permissive safeguards to vetted individuals and organizations. The docs list its status as active with verification required. It sits at the top of the Claude 5 lineup on price, ahead of Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, and Anthropic's comparison table marks its latency as slower than all three.
On Anthropic's own published evaluations, Fable 5.1 (and so Mythos 5.1, which shares its weights) scores 1,853 on GDPval-AA v2, a real-world knowledge-work benchmark, up from 1,723 for the previous Fable 5 generation. It scores 65.0% on Humanity's Last Exam with tools and 60.9% without, and Terminal-Bench-Science 0.1 rose from 24.7% to 52.6%. The ARC Prize foundation recorded 97.5% on ARC-AGI-1 at $1.40 per task and 90.0% on ARC-AGI-2 at $4.49 per task for Fable 5.1 at maximum effort. On BenchLM's SWE-bench Pro leaderboard, in its October 2026 snapshot, Fable 5.1 sits at 81.2%, behind Claude Opus 5.5 at 89.9% and Claude Sonnet 5.5 at 81.3%. Scores for SWE-bench Verified, GPQA Diamond, AIME and MMLU-Pro were not found in the sources read for this review, so those fields stay blank rather than estimated.
Context runs to 1 million tokens with a 128,000 token maximum output, and the reliable knowledge cutoff is June 2026. Thinking is adaptive and always on with a default effort of high. Input is text and images and output is text only, the same split Anthropic ships across Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. The ID is claude-mythos-5-1 on the Claude API and anthropic.claude-mythos-5-1 on Amazon Bedrock, with Google Cloud and Microsoft Foundry also listed. Anthropic lists retirement as not sooner than September 1, 2027.
Access is the defining difference. Anthropic's docs say Mythos 5.1 is available only to organizations verified through its verification programs, such as the Cyber Verification Program (CVP). In October 2026 Anthropic reworked the CVP into three access tiers, Defense, Red Team and Specialized, covering Claude Opus 5.5, Claude Sonnet 5.5 and Mythos 5.1. Organizations apply once through the Verification Portal and Anthropic aims to reply within seven business days; individuals can apply for Defense Access only. Organizations previously enrolled in Project Glasswing (see Project Glasswing explained) move to the Specialized tier without reapproval, and Specialized applicants are reviewed in depth with the US government. Anthropic says it is working to widen eligibility in the US and internationally. On cloud platforms, Mythos access trails approval by about five business days, and Amazon Bedrock is limited to customers with Enterprise Frontier Safeguards or a data retention exemption. For life sciences, Anthropic's September 1 announcement describes a separate Life Sciences Verification Program, built with the US government, with first participants enrolled. In Claude plans, Mythos 5.1 follows Fable 5.1's usage rules, and Pro plans need usage credits for it.
What verification unlocks is narrower than it sounds. Fable 5.1 itself is now allowed to identify software vulnerabilities for defensive work, and Anthropic says Claude Code users see around 60% fewer cyber safeguard interventions per session than on Fable 5. Fable 5.1 still redirects penetration testing, exploit generation and binary-based vulnerability scanning to Opus models. Anthropic says CVP exists for work such as malware analysis and exploit validation that its general classifiers may interrupt, and the scope depends on the tier approved. On biology, safeguards fire 85% less often for benign elementary biology and medical questions than at Fable 5's launch, while life-science research and development queries on Fable 5.1 still go to Opus models. New API accounts also cannot edit prior context while preserving the transcript of earlier thinking, an anti-distillation change Anthropic added with this release.
The other side of the argument is cost and fit. Mythos 5.1 costs 2.5 times Claude Opus 5.5 and 5 times Claude Sonnet 5.5 per token, and BenchLM lists Opus 5.5 ahead of the shared weights on SWE-bench Pro, so teams without a verified security or life-sciences need get more for less from Claude Opus 5.5 or Claude Sonnet 5.5. The generally available Claude Fable 5.1 carries the same weights and price with no application. Mythos 5.1 fits CVP members who need the relaxed classifiers and Life Sciences Verification Program participants running research workflows.
Data handling follows the program: CVP requires data retention so Anthropic can monitor for harmful cyber misuse, organizations with a data retention exemption for Fable or Mythos models can use it with zero data retention, and Enterprise Frontier Safeguards will let eligible organizations store data in infrastructure they control once available. API inputs and outputs are not used to train Anthropic's models by default. Mythos 5.1 and Fable 5.1 replaced Claude Mythos 5 and Claude Fable 5 on September 1, 2026, and Anthropic estimates Fable 5.1 costs about 25% less than Fable 5 for typical workloads, mostly from cheaper cache reads, with savings up to about 45% on highly agentic work.
Pricing
Input costs $10 per million tokens and output $50 per million, the same rate as [Claude Fable 5.1](/hub/models/claude-fable-5.1) since both are priced identically. A 5-minute prompt cache write costs $12.50 per million tokens, a 1-hour cache write $20 per million, and cache reads $0.25 per million, a 75% cut from the previous generation. The asynchronous Batch API gives a 50% discount on both input and output. There is no free tier, and access is limited to verified organizations. On Claude Pro plans Mythos 5.1 needs usage credits, and Max plans share an allowance of up to 50% of the weekly limit with Fable.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.300 | $0.050 | $0.350 |
| Support reply | $0.020 | $0.015 | $0.035 |
| One coding agent run | $2.00 | $1.00 | $3.00 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- 1 Million Token Context: Processes up to 1 million tokens of input, the same window Anthropic ships across the whole Claude 5 lineup, matching the output ceiling used across that lineup on the synchronous Messages API.
- Adaptive Extended Thinking: Reasoning effort adjusts per request under an always-on adaptive mode, defaulting to high unless a caller sets it lower to save tokens.
- Reduced-blocking Cyber Classifiers: Verified Cyber Verification Program members get reduced-blocking classifiers for security work that general-access models can interrupt, with the scope set by the Defense, Red Team or Specialized tier.
- Life Sciences Research Access: Life-science professionals can apply to Anthropic's Life Sciences Verification Program to use Mythos 5.1 with safeguards designed for research and development work, per the September 1 announcement.
- Multi-Platform Deployment: Model ID claude-mythos-5-1 works on the Claude API; Amazon Bedrock uses anthropic.claude-mythos-5-1, and Google Cloud and Microsoft Foundry are also supported.
- Batch API Support: Run non-urgent jobs through the asynchronous Message Batches API and pay half price on input and output tokens.
Pros
- Its strongest reasoning score comes from the independent ARC Prize foundation, measured on the shared Fable 5.1 weights, rather than from Anthropic's own marketing claims.
- Anthropic's own evaluations show large gains over the previous generation in knowledge work and agentic scientific research.
- Cybersecurity and biology safeguards are set for verified defensive and research work rather than left at Fable 5.1's general-audience defaults.
- Shares its 1 million token context window and pricing with the generally available Claude Fable 5.1, so nothing is held back once access is granted.
Cons
- Only reachable after a formal verification process, which rules it out for developers who just want to try a Claude model on a whim.
- What it unlocks depends on the tier Anthropic approves, and CVP grants cannot be used to build client-facing products without separate approval.
- Priced well above Anthropic's Opus and Sonnet models per token, and one of those models sits ahead of these weights on a leading agentic coding leaderboard.
- The Specialized tier is reviewed in depth with the US government, and broader eligibility is still being worked on.
Benchmarks
- ARC-AGI 2: 90% independent · 12 Oct 2026 — Abstract visual puzzles built to resist memorisation, % solved.
- GDPval-AA v2: 1,853 vendor-reported · 12 Oct 2026 — Real knowledge-work deliverables judged against professionals, run by Artificial Analysis.
- SWE-bench Pro: 81.2% independent · 12 Oct 2026 — Harder, longer real-repository coding tasks, % solved.
- Humanity's Last Exam: 65% vendor-reported · 12 Oct 2026 — Expert-written questions across many fields, % correct.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does Claude Mythos 5.1 cost?
Per-token rates match Claude Fable 5.1 exactly: $10 per million tokens in and $50 per million out. Writing a fresh prompt cache entry costs $12.50 per million for 5 minutes or $20 per million for 1 hour, while reading a cached entry costs $0.25 per million, far below the prior generation. The Batch API halves both input and output rates, and no free tier exists.
How do you get access to Claude Mythos 5.1?
Anthropic's docs restrict it to organizations that its programs have verified. For security work, apply once per organization through the Verification Portal for the Cyber Verification Program, which now has Defense, Red Team and Specialized tiers; Anthropic aims to reply within seven business days. Life-science professionals use the separate Life Sciences Verification Program.
What can you use instead of Claude Mythos 5.1 without verification?
Claude Fable 5.1 is the same model at the same price, with standard safeguards and no application. For general coding and reasoning work, Claude Opus 5.5 ($4 in, $20 out) and Claude Sonnet 5.5 ($2 in, $10 out) cost far less, need no vetting, and BenchLM lists Opus 5.5 at 89.9% on SWE-bench Pro against 81.2% for Fable 5.1.
Is Claude Mythos 5.1 better than Claude Fable 5.1?
They are the same underlying model with identical context window, output limit and pricing, so the choice is about safeguards and access. Pick Fable 5.1 for the generally available version, which now also lets you identify software vulnerabilities for defensive work. Pick Mythos 5.1 only once your organization is verified and needs the relaxed classifiers.
What changed in the Cyber Verification Program in October 2026?
The program now has three tiers named Defense, Red Team and Specialized, and they now apply to Opus 5.5, Sonnet 5.5 and Mythos 5.1. Members of Project Glasswing move to the Specialized tier without reapplying, individuals can apply for Defense Access only, and Defense Access must drop API keys and adopt phishing-resistant multi-factor authentication by December 15, 2026.
HokAI guides covering Claude Mythos 5.1
- Project Glasswing Explained: What Claude Mythos Found Inside Epic, and Who Else Gets Access: What Project Glasswing is, who is in it, what Epic found with Claude Mythos, and how Google Fairwind and OpenAI Codex Security compare on access and price.
- Which Claude Model Should You Use? Opus, Sonnet and Haiku by Job in 2026: Opus 5.5, Sonnet 5.5, Haiku 5.5 or Fable 5.1: which Claude model fits your job in 2026, with Anthropic's current prices, benchmarks and model guidance.