Claude Fable 5 vs Claude Mythos 5
Claude Fable 5 and Claude Mythos 5 share the same Mythos-class architecture, but they are not interchangeable products. At launch, Fable 5 was the general Claude route for production teams, while Mythos 5 was invite-only through Project Glasswing for approved cybersecurity defenders and biomedical researchers. Anthropic has since suspended both routes while it works to restore access, so this page separates launch capability evidence from current availability.
Do not pick either Claude Fable 5 or Claude Mythos 5 for new production work while access remains unavailable. When access returns, pick Fable 5 for normal production adoption across Claude API, Bedrock, Vertex AI, and Microsoft Foundry at the tracked $10/M input and $50/M output launch price. Track Mythos 5 only if you are eligible for Project Glasswing or a trusted access program. Do not attribute Mythos 5 starred benchmark scores, including Terminal-Bench 2.1 88.0% and HLE with-tools 64.5%, to Fable 5.
Decision scorecard
Local evidence first| Signal | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Best for | reasoning-heavy apps, multimodal apps, and tool-calling agents | reasoning-heavy apps, multimodal apps, and tool-calling agents |
| Decision fit | Coding, RAG, and Agents | Coding, RAG, and Agents |
| Context window | 1m | 1m |
| Cheapest output | $50/1M tokens | $50/1M tokens |
| Provider routes | 8 tracked | 3 tracked |
| Shared benchmarks | 0 shared | 0 shared |
Decision tradeoffs
- Claude Fable 5 has broader tracked provider coverage for fallback and procurement flexibility.
- Local decision data tags Claude Fable 5 for Coding, RAG, and Agents.
- Claude Mythos 5 uniquely exposes Code execution in local model data.
- Local decision data tags Claude Mythos 5 for Coding, RAG, and Agents.
Monthly cost at traffic
Estimate token spend from the cheapest tracked input and output route or tier on this page.
Claude Fable 5
$20,500
Cheapest tracked route/tier: Anthropic
Claude Mythos 5
$20,500
Cheapest tracked route/tier: Anthropic
Estimated monthly gap: $0.00. Batch, cache, alternate speed tiers, and negotiated pricing are excluded from this local estimate.
Switch friction
- Provider overlap exists on Anthropic, AWS Bedrock, and GCP Vertex AI; start route-level A/B tests there.
- Cheapest tracked output pricing is tied, so migration risk shifts to quality, latency, and provider packaging.
- Claude Mythos 5 adds Code execution in local capability data.
- Provider overlap exists on Anthropic, AWS Bedrock, and GCP Vertex AI; start route-level A/B tests there.
- Cheapest tracked output pricing is tied, so migration risk shifts to quality, latency, and provider packaging.
- Check replacement coverage for Code execution before moving production traffic.
Specs
| Specification | ||
|---|---|---|
| Released | 2026-06-09 | 2026-06-09 |
| Context window | 1m | 1m |
| Parameters | — | — |
| Architecture | Decoder Only | Decoder Only |
| License | Proprietary | Proprietary |
| Openness | Proprietary | Proprietary |
| Weights | Not released | Not released |
| Code | Unknown | Unknown |
| Commercial use | Commercial use: conditional | Commercial use: conditional |
| Knowledge cutoff | - | - |
Pricing and availability
| Pricing attribute | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Input price | $10/1M tokens | $10/1M tokens |
| Output price | $50/1M tokens | $50/1M tokens |
| Providers |
Capabilities
| Capability | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Vision | Yes | Yes |
| Multimodal | Yes | Yes |
| Reasoning | Yes | Yes |
| Function calling | Yes | Yes |
| Tool use | Yes | Yes |
| Structured outputs | Yes | Yes |
| Code execution | No | Yes |
| IDE integration | No | No |
| Computer use | No | No |
| Parallel agents | No | No |
Benchmarks
No shared benchmark scores are currently available for this pair.
Deep dive
The access model is the practical split. Claude Fable 5 launched as the broadly available model Anthropic wanted most developers to call, with listed pricing and mainstream provider routes. Claude Mythos 5 is source-backed as a distinct model, but access is restricted to approved Project Glasswing participants, with Bedrock and Vertex availability described as limited rather than self-serve. Both are currently marked unavailable in the seed.
The benchmark table needs a guardrail before any score comparison. Anthropic's Fable 5/Mythos 5 launch table shows the higher score between the two models. Rows marked with an asterisk are Mythos 5 scores, not Fable 5 scores. LLMReference seeds Fable 5 only with unstarred rows or independent third-party rows, so the Fable 5 page gets SWE-bench Pro 80.3%, Vals.ai SWE-bench Verified 96%, OSWorld-Verified 85.0%, and GDPval-AA ELO 1932.
Mythos 5 exists for restricted-domain work. Its starred Terminal-Bench 2.1 and HLE rows reflect the model with safety classifiers lifted for Glasswing contexts. Fable 5 routes sensitive biology, cybersecurity, and distillation requests through safer behavior, so copying Mythos-only scores onto Fable would overstate what a normal production caller should expect.
Pricing is similar on the first-party numbers but not equivalent in procurement risk. The tracked Anthropic rows list both Fable 5 and Mythos 5 at $10/M input and $50/M output. Fable 5 has normal listed routes once restored; Mythos 5 pricing is partner-listed and access-gated, so eligibility and provider approval matter before token math.
For most teams, the migration path is Opus or Sonnet to Fable 5, not Mythos 5, after access returns. Use Mythos 5 as a specialized restricted-access benchmark and safety-policy reference. Use Fable 5 for the actual model-provider decision unless your organization has explicit Glasswing approval.
FAQ
Is Claude Mythos 5 available to normal Claude API users?
No. Claude Mythos 5 is tracked as Project Glasswing invite-only access for approved cybersecurity defense and biomedical research organizations, and both Mythos 5 and Fable 5 are currently marked unavailable after Anthropic suspended access.
Can I use Mythos 5 benchmark scores for Claude Fable 5?
Do not copy starred Mythos 5 scores onto Fable 5. Anthropic's shared launch table reports the higher Fable-or-Mythos score, and starred rows are Mythos 5 only. Fable 5 should use only unstarred launch rows or independent Fable-specific measurements.
Which benchmark rows are Fable 5 evidence?
The seeded Fable 5 rows are SWE-bench Pro 80.3%, Vals.ai SWE-bench Verified 96%, OSWorld-Verified 85.0%, and GDPval-AA ELO 1932. Terminal-Bench 2.1 88.0% and HLE with-tools 64.5% are Mythos 5 rows.
Which should I choose for production?
Choose neither for new production work while both models are unavailable. When access returns, choose Claude Fable 5 for ordinary production work and consider Claude Mythos 5 only if your organization has Project Glasswing access and needs the restricted-domain safety-policy changes.
Continue comparing
Last reviewed: 2026-07-20. Data sourced from public model cards and provider documentation.