Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, reducing the cache-read price for Fable 5.1 by 75% — from $1.00 per million tokens to $0.25 per million — while keeping list prices unchanged at $10.00/M for input and $50.00/M for output tokens.
The cache-read cut reduces typical API spend by approximately 25% and cuts costs for highly agentic workloads — those with long tool-call chains and repeatedly reused large contexts — by up to 45%, according to Anthropic. Both models are available immediately on the Claude API, AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure.
Why Cache Pricing Is the Headline
Cache reads represent the dominant cost driver in agentic pipelines, where the same large context — system prompt, tool definitions, document corpus — is submitted to the model across many sequential calls. At $1.00/M, Fable 5’s cache-read rate had been the primary cost objection from developers building multi-step best AI agents for business tasks. The 75% cut to $0.25/M addresses that objection without Anthropic reducing its list prices.
Input pricing ($10.00/M) and output pricing ($50.00/M) are unchanged. The discount applies exclusively to cache reads — the reuse of previously processed context — not to freshly submitted tokens. Existing integrations using the updated model string (claude-fable-5-1) receive the lower rate automatically with no migration required.
The Fable 5.1 context window is 1 million tokens with a maximum output of 128,000 tokens. Knowledge cutoff is June 2026.
Always-On Adaptive Thinking: What Changed for Developers
Fable 5.1 enables always-on adaptive thinking — the model applies extended reasoning automatically based on task complexity. Fable 5 required a separate “extended thinking” mode flag, which forced developers to maintain two distinct prompting strategies for simple and complex tasks. Fable 5.1 removes that distinction: one model string, one configuration.
Anthropic describes Fable 5.1 as its “most advanced model for coding and knowledge work,” citing early evidence of AI contributions to scientific research.
Benchmark Scores: Fable 5.1 vs. Fable 5 and GPT-5.6 Sol
Fable 5.1 outperforms Fable 5, Opus 5, and GPT-5.6 Sol across all 7 benchmark categories measured at launch. The largest gain is on Terminal-Bench-Science: Fable 5.1 scores 52.6% versus Fable 5’s 24.7%, a 113% relative increase. This benchmark tests autonomous code execution in science-oriented tasks — a dimension that directly supports the “scientific progress” framing in Anthropic’s official announcement.
The Fable 5.1 launch coincides with OpenAI’s “Path to Astra” announcement, which targets cybersecurity frontier capabilities. Anthropic is positioning Fable 5.1 on a different competitive axis — coding and knowledge work — rather than contesting the cybersecurity dimension directly. Developer reception has been strong: the Hacker News launch thread reached a score of 1,075, ranking as the top story on the board at the time of publication.
Claude Mythos 5.1: Same Model, Restricted Access
Claude Mythos 5.1 shares the same underlying model weights as Fable 5.1 and is not publicly available. Anthropic restricts Mythos 5.1 access to vetted organizations conducting offensive cybersecurity research or advanced biomedical work, where Anthropic’s default production safeguards are reduced. The dual-release pattern — one model, two safeguard configurations — matches the Fable 5 and Mythos 5 release structure from earlier in 2026.
What This Means for Businesses Evaluating AI Tools
For businesses evaluating best AI tools for business in 2026, Fable 5.1’s cache pricing changes the cost calculus for document-heavy agentic workflows: contract analysis, policy review, customer support pipelines, and multi-step research summarization are all tasks where the same large context is queried repeatedly. A 45% reduction in agentic API spend is a material operating cost change for teams running these workloads at scale.
The benchmark gains and always-on adaptive thinking are meaningful incremental improvements, but the pricing architecture is the more consequential business decision in this release. Fable 5.1 became more price-competitive without Anthropic making a public list-price cut — a positioning move that lets Anthropic address developer cost objections while preserving the perception of its premium rate card.
For Context
witho2.com tracked the Fable 5 billing changes that preceded this launch. For the pricing history leading up to this cache-read cut, see: Fable 5 pricing history — usage credits explained (July 2026).
