Claude Mythos 5: Fable 5 Safeguards Lifted (2026)
Claude Mythos 5 by Anthropic: the same model as Fable 5 with cybersecurity/biology safeguards lifted, for vetted Project Glasswing partners. 1M context, $10/$50 per MTok. Restored to full availability July 1, 2026 after a brief export-control suspension.
Claude Mythos 5 is the same underlying model as Claude Fable 5 with its cybersecurity, biology/chemistry, and distillation safeguards lifted, available only through Project Glasswing to vetted infrastructure-security partners. It shares Fable 5's 1M context and 95.0% SWE-bench Verified score at the same $10/$50 per 1M token pricing. Like Fable 5, it was suspended June 12, 2026 under a US export-control directive and restored July 1, 2026.
Claude Mythos 5, released June 9, 2026 by Anthropic, is the Mythos-class configuration of Claude Fable 5 offered through Project Glasswing to vetted partners, sharing Fable 5's 1M context window and benchmark performance at the same $10/$50 per 1M token pricing. Like Fable 5, Mythos 5 was suspended June 12, 2026 under a US export-control directive following a disclosed jailbreak, and access was restored July 1, 2026 after Anthropic deployed an improved safety classifier blocking the disclosed bypass in over 99% of cases.
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 Anthropic's Mythos-class AI, released June 9, 2026 as a limited-availability configuration for vetted partners. It shares the exact same underlying model as Claude Fable 5, Anthropic's generally available flagship — the two are not separate architectures. The difference is safeguard configuration: Fable 5 ships with three classifier-based safeguards that automatically fall back to Claude Opus 4.8 for cybersecurity queries, biology and chemistry requests, and distillation (capability-extraction) attempts. Mythos 5 has these three safeguards lifted, giving it full, unrestricted access to the underlying model's capabilities in those sensitive domains. Access is not self-serve. Mythos 5 is available to current Project Glasswing partners — cybersecurity and critical-infrastructure organizations Anthropic vets and onboards directly — plus an incoming trusted-access program for biomedical researchers, in which only the cybersecurity safeguard stays active. There is no public signup; prospective partners go through Anthropic's vetting process. Project Glasswing partners deploying Mythos-class models have found more than 10,000 high- or critical-severity security vulnerabilities in real-world codebases, and Anthropic highlights it as the strongest cybersecurity capability of any Claude model. On the biology and chemistry side, Anthropic cites internal protein-design workflows seeing up to 10x acceleration, genomics researchers using Mythos 5 to autonomously assemble single-cell data across 138 species, and molecular-biology hypotheses generated by Mythos 5 being preferred roughly 80% of the time over Opus-class alternatives in expert review. Mythos 5 shares Claude Fable 5's general-purpose benchmark performance, since both run on the same underlying model: a 1M-token context window, 128K max output tokens, always-on adaptive thinking, and Fable 5's published scores on SWE-bench Verified, SWE-bench Pro, and MMLU-Pro. Anthropic has not published a dedicated public benchmark isolating Mythos 5's uplift on cybersecurity tasks specifically. Pricing is identical to Claude Fable 5 at $10 per million input tokens and $50 per million output tokens, with the same Batch API, prompt-caching, and data-retention terms. Like Fable 5, Mythos 5 was suspended on June 12, 2026 under a US export-control directive after a jailbreak affecting the shared underlying model was disclosed, and access was restored for Project Glasswing partners on July 1, 2026 after Anthropic deployed an improved safety classifier.
Pricing
$10/$50 per 1M input/output tokens (same as Fable 5; less than half the price of Claude Mythos Preview). Batch API: $5/$25 per 1M (50% off, async within 24h). Cache read: $1/MTok (0.1x input rate). 5-min cache write: $12.50/MTok; 1-hr cache write: $20/MTok. US-only inference (inference_geo: 'us') adds 1.1x multiplier on all token categories. Access requires Project Glasswing approval; pricing applies to approved partners only.
Key Features
- Cybersecurity, Biology, and Distillation Safeguards Lifted: The three classifier-based safeguards that route Fable 5 requests to Opus 4.8 -- cybersecurity, biology/chemistry, and distillation -- are lifted on Mythos 5 for vetted partners. Available only through Project Glasswing.
- 10,000+ Vulnerabilities Found by Glasswing Partners: Anthropic reports Project Glasswing partners deploying Mythos-class models have found more than 10,000 high- or critical-severity security vulnerabilities in real-world codebases.
- 1M Token Context Window: Full 1 million token context at standard pricing with no surcharge, identical to Fable 5, supporting complete codebase ingestion for security audits and large-scale literature analysis.
- Same Underlying Model as Claude Fable 5: Mythos 5 and Fable 5 run on the same base weights, so general-purpose benchmark scores (95.0% SWE-bench Verified, 91.5% MMLU-Pro) are identical between the two.
- Adaptive Thinking, Always On: Internal reasoning is always active on every request, with depth controlled via the effort parameter -- unchanged from Fable 5.
Pros
- [object Object]
- [object Object]
- [object Object]
- [object Object]
Cons
- [object Object]
- [object Object]
- [object Object]
Benchmarks
- gdppdf: 29.8
- mmlu pro: 91.5
- gdpval aa: 1932
- exploit bench: 78
- swe bench pro: 80.3
- swe bench verified: 95
- frontier code diamond: 29.3
- health bench professional: 66
Frequently Asked Questions
What is Claude Mythos 5 and how does it differ from Claude Fable 5?
Claude Mythos 5, released June 9, 2026, is the same underlying model as Claude Fable 5 -- they are not separate architectures. The difference is safeguard configuration: Fable 5 ships with three classifier-based safeguards that automatically fall back to Claude Opus 4.8 for cybersecurity queries, biology/chemistry requests, and distillation (capability-extraction) attempts. Mythos 5 has these three safeguards lifted, giving vetted users full access to the underlying model's capabilities in those domains. Fable 5 is generally available to any developer; Mythos 5 is limited to vetted Project Glasswing partners. On shared general-purpose benchmarks (SWE-bench Verified 95.0%, MMLU Pro 91.5%, SWE-bench Pro 80.3%), both models perform identically since they run on the same base weights. Anthropic has not published a dedicated public benchmark isolating Mythos 5's cybersecurity uplift over Fable 5.
How much does Claude Mythos 5 cost per 1M tokens?
Claude Mythos 5 costs $10 per million input tokens and $50 per million output tokens, the same price as Claude Fable 5. Cache reads cost $1 per million tokens (0.1x the input rate); 5-minute cache writes cost $12.50 per million (1.25x) and 1-hour writes cost $20 per million (2x). The Batch API delivers a 50% discount: $5 input and $25 output per million, with results returned within 24 hours. US-only data routing using inference_geo: 'us' adds a 1.1x multiplier on all token categories. Pricing applies only to approved Project Glasswing partners; standard API accounts do not have access to the claude-mythos-5 model ID. Self-hosting is not available -- Mythos 5 is closed-weights and API-only.
What is Claude Mythos 5's context window and max output?
Claude Mythos 5 supports a 1 million token context window, identical to Fable 5, with up to 128,000 output tokens per request. The full 1M context is billed at the standard per-token input rate with no long-context surcharge. Prompt caching applies across the full context. The native memory tool is supported, enabling agents to store findings across conversation boundaries in long research sessions. Anthropic has not published independent needle-in-haystack evaluations for the full 1M range specific to Mythos 5.
What does Claude Mythos 5 do differently in cybersecurity?
With its cybersecurity safeguard lifted, Mythos 5 can engage with security research tasks that Fable 5's classifier routes to Opus 4.8 instead, such as vulnerability analysis on live codebases. Anthropic describes Mythos-class models as having the strongest cybersecurity capability of any Claude model, and Project Glasswing partners deploying Mythos-class models on real infrastructure have collectively found more than 10,000 high- or critical-severity security vulnerabilities. Anthropic has not published a standardized public benchmark score (such as an "ExploitBench") isolating this capability, and independent third-party verification of the real-world vulnerability-finding figures has not been published in detail. On shared general benchmarks, Mythos 5 matches Fable 5: 95.0% SWE-bench Verified, 80.3% SWE-bench Pro, 91.5% MMLU Pro.
Who can access Claude Mythos 5 and how do you apply?
Claude Mythos 5 is available exclusively through Project Glasswing, Anthropic's program for vetted cybersecurity and critical-infrastructure organizations, plus an incoming trusted-access program for biomedical researchers in which only the cybersecurity safeguard stays active. There is no self-serve enrollment: prospective partners go through Anthropic's own vetting and onboarding process. Organizations that do not qualify, or are waiting on approval, should use Claude Fable 5, which delivers identical general-purpose benchmark performance. Once approved, Mythos 5 is accessible via the Anthropic API using the model ID claude-mythos-5.
What modalities does Claude Mythos 5 support?
Claude Mythos 5 accepts text, images, PDFs, and tool calls as inputs, and produces text and tool calls as outputs -- identical modality support to Claude Fable 5. Vision is fully live: the model reads charts, interprets screenshots, and extracts data from PDF figures. Audio and video are not supported. Function calling is native with parallel tool execution. The key difference from Fable 5 is behavioral (which requests get refused versus routed to Opus 4.8), not input/output format.
Does Claude Mythos 5 train on user data?
Anthropic states it does not train on API inputs from approved Glasswing partners. All Mythos 5 sessions carry mandatory 30-day data retention for safety monitoring, which cannot be opted out of on any platform -- this mirrors Fable 5's data-retention policy. Anthropic says retained data is used exclusively for safety monitoring, not model training, and all human access to retained data is logged.
Who is Claude Mythos 5 best for, and who should use Fable 5 instead?
Claude Mythos 5 is for vetted Project Glasswing partners: cybersecurity and critical-infrastructure organizations running defensive vulnerability research, and biomedical researchers with trusted-access approval. Any team that has not been approved through Glasswing should use Claude Fable 5 instead, which delivers identical general-purpose performance (same SWE-bench, MMLU Pro benchmarks) without the restricted-access requirement. Teams needing zero data retention should look at Claude Opus 4.8, since neither Fable 5 nor Mythos 5 offers a zero-retention option.