Claude Fable 5.1

• 📏 1M-token context window, up to 128K output tokens per response.
• 🧠 Always-on adaptive thinking with selectable effort levels.
• 🔧 Gains focus on long-running agentic coding, refactors, migrations, debugging.
• 📚 Stronger multistep research and document, spreadsheet and slide work.
• 🎯 Anthropic reports a partner's RedlineBench rising 47.9 to 57.0 versus Fable 5.
• 👁️ Multimodal input, function calling and web search per catalog capabilities.
• 🔒 Breaking changes: forced tool use errors; edited turns invalidate thinking blocks.
• 🏢 Released August 2026 alongside Anthropic's Mythos 5.1 variant.

Anthropic

Anthropic PBC is an American artificial intelligence company headquartered in San Francisco. Structured as a public benefit corporation, the lab develops large language models under the Claude name, with a research emphasis on building reliable, steerable,…

Claude Fable 5.1 is Anthropic's August 2026 point release in the Fable line, aimed at demanding reasoning and long-horizon agentic work. Anthropic's platform documentation suggests starting most workloads on Claude Opus 5 and moving to Fable 5.1 when evaluations at higher effort settings still fall short. It ships alongside Mythos 5.1, which Anthropic describes as the same underlying model served under different safeguard levels.

Against its direct predecessor Claude Fable 5 — now marked legacy in Anthropic's docs — the company frames 5.1 as an across-the-board improvement whose margin widens at higher effort levels, with gains concentrated in agentic coding across hours-long sessions (multi-file features, large refactors and migrations, debugging, code review), multistep research, and document, spreadsheet and slide production. Anthropic's launch post cites a customer's contract-redlining benchmark, RedlineBench, moving from 47.9 to 57.0 relative to Fable 5, attributed mainly to first-turn quality and more concise edits. Anthropic also publishes cost-versus-score curves on Terminal-Bench 4.0 and Terminal-Bench-Science 0.1 across effort settings.

Migration is not fully drop-in. Per Anthropic's release notes, forced tool use now returns an error, earlier models cannot read this model's reasoning traces, editing prior turns invalidates them, and agent loops may see fewer batched tool calls per turn than with Fable 5.

The capability envelope matches Fable 5 broadly: a 1M-token context, 128K maximum output, always-on adaptive thinking, vision, function calling and web search, with plans and summaries that tend to be more concise. Anthropic says it is working to bring these improvements to other lines, such as Claude Sonnet 5.