Claude Fable 5.1: Benchmarks, Pricing & Access (2026)

On September 1, 2026, Anthropic announced Claude Fable 5.1 and Claude Mythos 5.1. Input/output pricing holds at $10 / $50 per 1M tokens, while cache-read pricing drops to $0.25 per 1M tokens (a 75% cut). The API model ID is claude-fable-5-1, available via Claude.ai, the API, Claude Code, and AWS / Google Cloud / Microsoft Azure.

Like Claude Fable 5 and Mythos 5 from June 2026, Claude Fable 5.1 is the publicly available version with safeguards intact. Claude Mythos 5.1 is the same base model with relaxed safeguards, offered through a US-only access program for vetted cybersecurity and life-sciences experts. Both are framed as a minor version bump (5 → 5.1) rather than the structural tier change Anthropic introduced in June.

Source: Anthropic's official announcement

The official announcement highlights improvements in:

- Coding and knowledge work
- Reasoning on long-running tasks (accuracy across multi-step, sustained agentic execution)
- Scientific research (molecular design and computational biology)
- Task verification for agentic work (checking its own outputs)
- Writing quality

There is no deprecation announcement for the prior Fable 5 / Mythos 5 models. Existing workflows don't need to switch immediately — teams can evaluate the benchmark and cost impact before migrating.

These are the benchmark scores published in the official announcement. The largest gains are on Terminal-Bench-Science and Terminal-Bench 4.0, pointing to stronger performance on long-running scientific and terminal-driven agentic tasks.

The context window length is not stated in the official announcement. If you're designing around token limits, check the latest Claude API Docs before committing to a figure.

Per-token input/output pricing is unchanged from Fable 5, but the sharp cut to cache-read pricing lowers real-world costs. Anthropic states this yields roughly 25% savings on typical workloads and up to 45% on agentic use cases. Tools that lean heavily on prompt caching, such as Claude Code, stand to benefit the most.

Claude Fable 5.1 is available through Claude.ai (model picker), the API (specify model ID claude-fable-5-1), Claude Code, and all three major clouds — AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure. Existing code and prompts can typically migrate by swapping the model ID.

Claude Mythos 5.1 is not generally available. It's offered only through a vetted access program for cybersecurity and life-sciences experts, limited to US organizations. The announcement does not detail the application or vetting process, so eligible organizations should contact Anthropic directly.

On paper, Fable 5.1 reads as a near price cut disguised as a version bump — same per-token rate, better benchmarks, and a much cheaper cache-read tier. For agentic workloads that lean on caching, the total-cost impact can be significant even without a headline price change. It's also worth revisiting how Fable 5.1 fits alongside other tiers like Claude Opus 5 and Claude Sonnet 5 now that the cost structure has shifted.