Publication date: September 2, 2026
Anthropic released Claude Fable 5.1 on September 1, alongside Claude Mythos 5.1 for vetted users. Fable 5.1 is available across Anthropic and major cloud platforms under the API identifier claude-fable-5-1.
The most practical change is not only higher model capability. Anthropic cut cache-read pricing to $0.25 per million tokens, 75% below the previous level, which can materially reduce the cost of long-running agents that repeatedly reuse context.
What is Claude Fable 5.1 and what changed?
Fable 5.1 is the new generally available version of Anthropic’s Mythos-class model. Claude Mythos 5.1 uses the same underlying model but applies more permissive safeguards for selected cybersecurity and life-sciences users.
Anthropic says Fable 5.1 improves reasoning, coding, computer use and scientific work over Fable 5. The company also says lower effort settings can match or beat its predecessor on several evaluations, which matters for teams balancing latency, cost and quality.
The release also changes safeguard behavior. Anthropic says benign biology prompts trigger its biology protections 85% less often than at the launch of Fable 5. In cybersecurity, Claude Code users should see about 60% fewer safeguard interventions per session. Penetration testing, exploit generation and some binary vulnerability work can still be redirected to Opus models.
Claude Fable 5.1 pricing, access and benchmarks
| Item | Anthropic release detail |
|---|---|
| API model | claude-fable-5-1 |
| Standard input | $10 per million tokens |
| Standard output | $50 per million tokens |
| Cache reads | $0.25 per million tokens |
| Availability | Anthropic, AWS, Google Cloud, Microsoft Azure |
| Mythos 5.1 | Trusted access only |
| What is measured | Claude Fable 5.1 | GPT-5.6 Sol |
|---|---|---|
| Terminal-Bench-Science 0.1 (agentic science research) | 52.6% | 22.4% |
| Terminal-Bench 4.0 (agentic terminal coding) | 55.8% | 37.3% |
| CursorBench 3.2.0 (real coding tasks) | 73.4% | 67.2% |
| AutomationBench (business workflows) | 31.4% | 19.6% |
| GDPval-AA v2 (knowledge work, higher is better) | 1853 | 1711 |
| OSWorld 2.0, strict scoring (computer use) | 41.7% | no figure published |
These are vendor-reported results, not independent testing performed for this article. Anthropic publishes evaluation notes and settings alongside its benchmark table, but model and harness differences still make cross-vendor comparisons imperfect.
Who should test Fable 5.1?
Product engineering teams running coding agents are the clearest audience. Lower cache-read costs matter when an agent repeatedly loads a large repository, specification set or operating context. Security teams can also test whether the more precise defensive safeguards reduce unnecessary interruptions.
The main limitation is price. Standard Fable 5.1 input and output rates remain $10 and $50 per million tokens, so the release is still positioned above lower-cost general-purpose models. Mythos 5.1 is not a normal commercial option because access remains restricted to vetted programs.
DataNorth previously covered the original Claude Fable 5 and Mythos 5 launch, so this article focuses on the 5.1 release and its new pricing, performance and access details.
What this means
Teams already using Fable 5 should test Fable 5.1 on expensive long-context agent workflows first. Cache-heavy coding, document analysis and multi-step research are the areas most likely to show an immediate cost benefit.
Verdict: Fable 5.1 is definitely worth testing now. The combination of a new model, broader cloud availability and much cheaper cache reads creates a concrete migration case, although teams should validate Anthropic’s benchmark gains on their own workloads.
For more information, visit the official announcement of Claude Fable 5.1 on the Anthropic website.
