Anthropic: claude-fable-5-1

Claude Fable 5.1 is Anthropic's flagship model, released on September 1, 2026. It accepts text and image input and returns text output, with a 1-million-token context window and support for up to 128K tokens of output. Thinking is always adaptive, with the default effort level set to high. It is generally available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. The model is built for demanding reasoning, long-horizon agentic work spanning multiple days, scientific discovery, and enterprise workflows involving documents, spreadsheets, and slides. It also handles implementing features across an entire codebase, code review, and performance improvement. Anthropic reports state-of-the-art performance in coding and knowledge work, with Terminal-Bench 4.0 rising from 42.0% on Claude Fable 5 to 55.8%. It also scores 35.03% on RSI-Index, ranking #1, and its writing style is described as more natural and less stereotyped than earlier Claude models. Some trade-offs are worth noting. Output speed is 70.1 tokens per second, which is on the slower side, and the model is classified as relatively high-latency. The reported refusal rate is substantially higher than earlier Claude models, though cybersecurity-related refusals are reportedly down about 60%. Migrating from Fable 5 involves 4 breaking changes, including forced tool use now returning errors. The training data cutoff is June 2026, and retirement is not scheduled before September 1, 2027.