Claude Fable 5.1 is Anthropic's frontier model, built for ambitious coding projects, long-horizon agents and dense document work. Its context window reaches one million tokens. Shipped on September 1, 2026 alongside its restricted twin Mythos 5.1, it replaces Fable 5 from three months earlier and trims typical running costs by roughly 25%. Access goes through paid Claude plans or the token-billed API.
Fable 5.1 is designed for jobs that stretch over hours and span several applications, from clearing a backlog in Claude Cowork to driving a browser or picking up requests in Slack through the Claude Tag beta. You hand it a ticket list in the morning; it plans the work, recovers when a step fails and sends progress updates on its own.
The story making the rounds comes from the investment firm Millennium. A crash occurred about once in a million runs and had gone unexplained for years; Fable 5.1 traced it to a bug in an external library after every other model came up empty (it took the library apart piece by piece to get there). In Cursor, it also posted the highest score ever recorded on CursorBench 3.2, a fiercely contested arena for AI coding assistants.
On Terminal-Bench-Science, a test of autonomous scientific research, the score jumps from 24.7% to 52.6%, over double Fable 5's result. Anthropic also places the model ahead of Opus 5 and OpenAI's GPT-5.6 Sol on several coding and reasoning measures.
Vision got a quieter upgrade too. The model reads charts and tables buried inside PDFs, a relief for finance and legal work. And the effort level, from low through max, can now be adjusted mid-conversation.
| Benchmark | Fable 5 | Fable 5.1 |
|---|---|---|
| Terminal-Bench-Science 0.1 | 24.7% | 52.6% |
| Terminal-Bench 4.0 | 42.0% | 55.8% |
| CursorBench 3.2 (max effort) | not published | 73.4% |
Access runs through the Claude Pro, Max, Team and Enterprise plans, or through the API under the claude-fable-5-1 identifier. Expect $10 per million input tokens and $50 per million output tokens, unchanged from Fable 5; the real news hides in cache reads at $0.25 per million, a 75% drop that lowers real bills by about 25% on typical workloads and up to 45% on heavy agent runs. Anthropic's official pricing page remains the reference should these figures move.
The model is also served on Amazon Bedrock, Google Cloud, Microsoft Foundry and inside Microsoft Copilot for eligible organizations. One habit worth picking up early is starting at low or medium effort, where Fable 5.1 already matches its predecessor at a much lower cost.
No, Fable 5.1 requires a paid Claude plan (Pro, Max, Team or Enterprise) or an API account billed by the token. Claude's free tier does not include it. Subscribers get it at no extra charge, within the usage limits attached to their plan.
They are the same model with different safeguard levels. Fable 5.1 is generally available, while Mythos 5.1 keeps its full cybersecurity and biology capabilities and stays limited to vetted organizations in Anthropic's trusted access programs, currently U.S.-based. A verification program for life scientists has been announced.
Blocking classifiers filter sensitive cybersecurity and life-sciences queries, and refusal rates run higher than on earlier Claude models. Anthropic has cut security false positives in Claude Code by around 60%, and the model may hunt for software vulnerabilities but will not write exploits for them.
On Anthropic's published benchmarks, Fable 5.1 comes out ahead of OpenAI's GPT-5.6 Sol, along with Fable 5 and Opus 5, across several coding and reasoning tests. Its strongest card is long autonomous work, where it checks its own output from start to finish.
Verdict: A one-in-a-million bug cracked after years of dead ends says a lot about where this model belongs. Development teams and document-heavy professions will get the most from it, with Sonnet and Opus covering the lighter day-to-day questions.
Claude Fable 5 
Grok 4.6 
Claude Opus 4.8 
