Claude Fable 5.1 goes public while Mythos 5.1 stays behind trusted access
Anthropic's new shared model core brings stronger long-running coding and research to the public Fable tier, while Mythos 5.1 remains restricted and cheaper cache reads change the economics of agentic workloads.
Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. The two names point to the same underlying model with different safeguards: Fable 5.1 is generally available, while Mythos 5.1 is limited to Anthropic's trusted-access programs for vetted cyberdefenders and life-science professionals. 12
What launched
| Signal | Confirmed detail | Why it matters | Action window |
|---|---|---|---|
| One model, two access tiers | The API identifiers are claude-fable-5-1 and claude-mythos-5-1. Both support a 1M-token context window, up to 128K output tokens, and always-on adaptive thinking. Fable is available to all customers; Mythos is restricted to Project Glasswing participants. 3 | The capability comparison is shared, but the deployment and safeguard conditions are different. | Start general-purpose evaluation with Fable; treat Mythos as a controlled-access track. |
| Long-running work | Anthropic positions Fable 5.1 for long-running agentic coding, multistep research, and document, spreadsheet, and slide work. The model's effort setting lets applications trade thinking depth against latency and cost. 3 | The upgrade targets workflows that must preserve context and verify several steps, rather than short chat answers. | Run task-level evals with the same tools, prompts, and data used in production. |
| Lower cache cost | Input and output pricing stays at $10 and $50 per million tokens. Cache reads fall to $0.25 per million tokens, or one quarter of the previous price. Anthropic estimates roughly 25% lower cost for typical workloads and up to 45% for highly agentic workloads. 2 | Context-heavy agents may see a larger saving than ordinary requests, but the percentage depends on cache reuse. | Measure cost per completed task, not only token prices, before changing model routing. |
| Safeguards and integration limits | Fable 5.1 can identify software vulnerabilities, while penetration testing, exploit generation, and binary-based vulnerability scanning are redirected to Opus. Forced tool selection with tool_choice: "any" or a named tool returns an error; use automatic tool choice or structured outputs instead. 23 | Existing coding agents may need changes to tool orchestration and cybersecurity workflows. Fable 5.1 thinking blocks also bind to the model that produced them, which affects model fallback and history editing. 3 | Check tool-choice settings, fallback routes, and replayed conversation history before rollout. |
What the numbers say
Anthropic reports Fable 5.1 at 52.6% on Terminal-Bench-Science 0.1, 55.8% on Terminal-Bench 4.0, 1,853 on GDPval-AA v2, and 73.4% on CursorBench 3.2.0. The same table lists 60.9% for Mythos 5.1 on Terminal-Bench 4.0. These are Anthropic's evaluations, and the company says production safeguards intervened on some tasks, which can reduce the reported scores. The Terminal-Bench-Science comparison also carries a standard error of roughly 3.5–4.5 points per model. 2
Why this matters
The release separates a model's capability from the conditions under which a customer may use it. Fable 5.1 makes frontier-level coding and research available through the public API, with lower cache-read costs that could favor long-context agents. Mythos 5.1 gives a narrower group access to the same core model under more permissive cyber and life-science safeguards. 2
For practitioners, the immediate test is a controlled migration from Fable 5: replay real tasks, compare completion quality and wall-clock cost, then inspect tool-call behavior and safety redirects. A benchmark lead is a reason to evaluate the model; the access tier, safeguard routing, and integration changes determine whether the release fits a live system.
Fuentes de referencia
- 1Anthropic release notes
docs.anthropic.com
- 2Introducing Claude Fable 5.1 and Claude Mythos 5.1
anthropic.com
- 3Claude Platform model guide
docs.anthropic.com
Este contenido lo produjo un canal automáticamente. Con una sola frase, Neodrop puede seguir produciendo para ti.