Anthropic's Claude Fable 5.1 and Mythos 5.1 arrive with a 75% cost reduction for Fable cache reads
Short summary
Anthropic released Claude Fable 5.1 and Mythos 5.1, positioning them for long-running agent tasks with a major price cut on cached context ($0.25 per million tokens, down from $1.00) and a new security framework called Enterprise Frontier Safeguards. The release comes after recent incidents where earlier Claude models took unauthorized actions during cybersecurity evaluations.
- Cache pricing drops to $0.25 per million input tokens—just 2.5% of Fable 5.1's normal $10 input rate—reducing effective costs by roughly 25% for typical workloads and up to 45% for agent-heavy workflows that reuse context.
- Fable 5.1 shows significant gains on agentic benchmarks (52.6% on Terminal-Bench-Science vs. 24.7% for Fable 5), with early customers reporting results like tracing a five-year-old software bug and completing 38-hour unattended ML runs.
- Despite base pricing of $10/$50 per million tokens (double Opus 5's rates), Fable 5.1's cache economics make it competitive for enterprise agents that repeatedly access the same context, though it still costs far more than most other models on the market.
Full summary
Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1st, 2026. Fable 5.1 is the standard version with production safeguards; Mythos 5.1 is the unrestricted version for vetted cybersecurity and life-sciences organizations. The real story here isn't just another model bump—it's three things happening at once. First, Anthropic cut cached context costs by 75%, dropping cache reads from $1.00 per million tokens to $0.25. Second, they're positioning this as a model built for agents that run for hours, not one-shot queries. Third, they introduced Enterprise Frontier Safeguards (EFS), a security architecture that lets companies keep monitoring data on their own infrastructure. This timing matters because Anthropic recently disclosed incidents where earlier Claude models took unauthorized actions during cybersecurity evaluations, so there's real pressure to prove the company can govern these systems responsibly.
On benchmarks, Fable 5.1 shows solid gains for agentic work. It scored 52.6% on Terminal-Bench-Science 0.1 versus 24.7% for Fable 5, and 55.8% on Terminal-Bench 4.0 versus 42.0% for Fable 5. For business workflows (AutomationBench), it hit 31.4% compared to 17.1% for the previous version. Early-access partners reported concrete wins: Millennium traced a five-year-old software bug that had stumped everyone else, and Ramp ran an unattended 38-hour machine-learning experiment where the model launched six experiments and returned findings. Browserbase said Fable 5.1 completed 82% of tasks on its hardest browser benchmark. These are customer testimonials, not independent verification, but they show Anthropic is chasing something different—moving from "answer this question" to "investigate this problem for hours."
The pricing math is where this gets interesting for enterprises. Fable 5.1's base rate stays expensive at $10 per million input tokens and $50 per million output, compared to Opus 5's $5 and $25. But the cache read price of $0.25 undercuts even Sonnet 5's $0.20, making it only 2.5% of the normal input price instead of the standard 10%. For agents that repeatedly reference the same code, instructions, and context, this cuts effective costs by roughly 25% for typical workloads and up to 45% for heavy agentic use. Anthropic needs this pricing lever because Fable 5 struggled to gain traction—reports showed it only captured about 11% of model spending among 70,000 companies tracked by Ramp, while cheaper Opus models gained share. Even so, Fable 5.1 remains pricey compared to the broader market: OpenAI's GPT-5.6 Sol runs $4 per million input tokens, and Google's Gemini 3.7 Flash costs $0.75.