Claude Mythos 5 is Anthropic's verification-only version of Claude Fable 5, released on 9 June 2026 with cybersecurity safeguards lifted for vetted defenders. Since 1 September 2026 a newer sibling has been the current Mythos model, and Anthropic lists the older one as active until at least 9 June 2027.
What changed
Claude Mythos 5 is a restricted Anthropic language model released on 9 June 2026. It runs the same underlying model as Claude Fable 5, with a 1M-token context window, but is offered only to organisations verified through Anthropic's Cyber Verification Program, which absorbed Project Glasswing on 6 October 2026.
Where it sits
- $20.00/M$ per 1M tokensBlended price (3:1)Lower is better#83 / 89peer median $1.71/Mvendor price, checked by HokAI
- 95%% solvedSWE-bench VerifiedHigher is better#3 / 38peer median 76.8%per source, see benchmark scores
Pricier than 98% of the 89 GA models with a published price, in the top third on SWE-bench Verified (rank 3 of 38), and one of 33 that document a zero-data-retention option. Ranked against GA models; this record is not GA.
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, pdf, tool-calls · Output: text, tool-calls
About Claude Mythos 5
Claude Mythos 5 is the restricted-access configuration of Claude Fable 5, released by Anthropic on 9 June 2026 as the upgrade for partners already using Claude Mythos Preview in Project Glasswing. Anthropic describes the pair as one underlying model with different safeguards: Fable 5 passes cybersecurity, biology and chemistry, and distillation requests to Claude Opus 4.8, while Mythos 5 has its cybersecurity safeguards lifted for vetted defenders. A separate trusted-access program was planned to lift the biology and chemistry safeguards for selected life-science researchers.
The June launch was short-lived in practice. On 12 June 2026 the US government applied export controls to both models, and Anthropic suspended access for all users because it had no reliable way to verify nationality in real time. The controls were lifted on 30 June. Fable 5 returned on 1 July, and Mythos 5 was restored to a set of US organisations after government approval on 26 June.
Mythos 5 is no longer the current Mythos model. Claude Mythos 5.1, released on 1 September 2026, is the same model as Claude Fable 5.1 with trusted-access safeguards, and Anthropic's docs point new work at it. Anthropic still lists Mythos 5 as active, with retirement no sooner than 9 June 2027, so existing integrations keep working. On 6 October 2026 Anthropic folded Project Glasswing into an expanded Cyber Verification Program with Defense Access, Red Team Access and Specialized Access tiers. Existing Glasswing members move to Specialized Access and need no fresh approval for current models.
Mythos 5 shares Fable 5's 1M-token context window and 128K output ceiling, takes text and images, returns text, and always runs adaptive thinking. Access is by verification only. Anthropic requires 30-day retention for Mythos-class traffic and says it does not use that data to train Claude models. Partners reported at least 129,000 verified vulnerabilities across Mythos models between April and July 2026, a lower bound Anthropic built from partial survey data. General-purpose benchmark results match Fable 5 because the base model is shared, and the models leaderboard shows how that compares with other labs' models.
Who should not choose it: almost everyone. Without verification Mythos 5 cannot be called at all, and for teams that do qualify, Mythos 5.1 is the sensible default on a new build because it is the same model family with a newer knowledge cutoff and cheaper cache reads. Teams that need code review, patching of known issues or alert triage can stay on Fable 5.1, which Anthropic says can now find vulnerabilities in code you own but not develop exploits for them. Mythos 5 also has no fallback model catching sensitive requests, so its output needs application-layer controls before it reaches end users. If you are unsure which Claude fits, read which Claude model to use, browse other AI security tools, or let Smart Match shortlist options for your stack. Other Anthropic options on HokAI include the Opus-class Opus 5 model, a tier below Mythos-class models, and the Claude assistant, and the model directory lists the rest.
Pricing
Input tokens for Claude Mythos 5 cost $10 per million and output tokens $50 per million from [Anthropic](/hub/companies/anthropic), the same rate as [Claude Fable 5](/hub/models/claude-fable-5) and less than half the price of Claude Mythos Preview. Cache reads cost $1 per million tokens, 5-minute cache writes $12.50 and 1-hour writes $20. Batch API jobs are discounted 50% on both input and output, with results returned within 24 hours. US-only inference (inference_geo 'us') adds a 1.1x multiplier on all token categories. These rates are open only to customers that have cleared one of Anthropic's verification programs. The newer Mythos model keeps those rates but cuts cache reads to $0.25 per million tokens, so it is cheaper for context-heavy agent work. The [models leaderboard](/directory/models/leaderboard) compares token prices across labs.
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
- Cybersecurity safeguards lifted for verified organisations: Mythos 5 has the cybersecurity safeguards that send Fable 5 requests to a fallback model switched off, for organisations Anthropic has verified through its Cyber Verification Program.
- 129,000+ vulnerabilities found by Glasswing partners: Anthropic says partner reports cover at least 129,000 verified vulnerabilities found with Mythos models from April through July 2026, plus more than 33,000 rated critical or high severity across partner and Anthropic scans; partial data understates the total.
- 1M token context window: A 1 million token context and 128K token output ceiling, identical to Fable 5, supports whole-codebase ingestion for security audits.
- Same underlying model as Claude Fable 5: Mythos 5 and Fable 5 share one base model, so general-purpose scores such as 95.0% SWE-bench Verified and 91.5% MMLU-Pro are the same on both.
- Adaptive thinking, always on: Internal reasoning runs on every request, with depth controlled through the effort parameter, which defaults to high.
- Superseded by its successor since 1 September 2026: The successor model moves the knowledge cutoff from January to June 2026, while Anthropic lists this one as retiring no sooner than 9 June 2027.
Pros
- Cybersecurity safeguards are lifted, so verified teams can run vulnerability research that Fable 5 hands to a fallback model.
- Several Glasswing partners told Anthropic that Mythos models raised their vulnerability-finding rate by months or even years.
- A 1M token context lets an agent hold a full codebase in one window, backed by always-on adaptive thinking.
- General-purpose quality matches Fable 5, so lifting safeguards does not cost reasoning or coding strength.
Cons
- Access runs through Anthropic verification, with no self-serve signup. Use the generally available Fable model while waiting for approval.
- A successor is the current Mythos model, with a fresher knowledge cutoff and lower cache-read rates, so new builds should target it.
- Safeguards are lifted and no fallback model screens requests, so applications must add their own filtering for anything user-facing.
- Retention of 30 days is mandatory for Mythos-class traffic, and Anthropic's October 2026 program post names zero data retention only for its newer models.
Benchmarks
- GDP-PDF: 29.8 vendor-reported · 09 Jun 2026
- MMLU-Pro: 91.5% vendor-reported · 09 Jun 2026 — A harder version of the 57-subject knowledge exam, % correct.
- GDPval AA: 1,932 vendor-reported · 09 Jun 2026
- Exploit Bench: 78 vendor-reported · 09 Jun 2026
- SWE-bench Pro: 80.3% vendor-reported · 09 Jun 2026 — Harder, longer real-repository coding tasks, % solved.
- SWE-bench Verified: 95% vendor-reported · 09 Jun 2026 — Real GitHub issues fixed end to end, % solved.
- Frontier Code Diamond: 29.3 vendor-reported · 09 Jun 2026
- Health Bench Professional: 66 vendor-reported · 09 Jun 2026
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
What does Claude Mythos 5 actually cost?
Anthropic charges the same as for Claude Fable 5: $10 per million input tokens and $50 per million output. Cache reads cost $1 per million, 5-minute cache writes $12.50 and 1-hour writes $20, and the Batch API takes 50% off input and output. Access is by verification only, so these prices apply to approved organisations, and Mythos 5.1 charges $0.25 per million for cache reads.
Who can access Claude Mythos 5?
Only organisations that Anthropic has verified. Its model page names the Cyber Verification Program as the route. That program now holds Project Glasswing inside three tiers: Defense Access, Red Team Access and Specialized Access. Existing Glasswing members move to Specialized Access without re-approval for current models; unverified teams use Claude Fable 5.1.
Which tools compete with Claude Mythos 5 in 2026?
The closest options come from the same family. Claude Mythos 5.1 is the newer verified-only model and the better pick for new builds. Claude Fable 5.1 suits teams with no need to lift a safeguard, since it now finds vulnerabilities in code you own. Claude Opus 4.8 is where Fable hands flagged requests, which makes it the fallback to compare against.
What separates Claude Mythos 5 from Claude Mythos 5.1?
Anthropic calls Mythos 5.1 the current Mythos model. It was released on 1 September 2026, stops its knowledge in June 2026 where Mythos 5 stops in January, and cuts cache-read pricing by 75%. Context and output limits are unchanged. Anthropic lists both as active, with Mythos 5 retiring no sooner than 9 June 2027, so a new integration should target Mythos 5.1.
How do you get started with Claude Mythos 5?
Apply to Anthropic's Cyber Verification Program through its portal. Anthropic says Defense Access applications get a reply within a few days, Red Team Access takes a few weeks, and Specialized Access involves in-depth review with the US government. Once approved, an admin assigns the program to workspaces in the Claude Console and the model is called as claude-mythos-5. Teams waiting on approval can prototype on Fable 5.1.
HokAI guides covering Claude Mythos 5
- Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5: Cost Per Real Task: Argon, GPT-6 Astra and Claude Opus 5.5 compared on what a finished task costs, not the token price, plus the 1 December price rise on Argon. Updated Oct 2026.
- 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.