Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, marking a meaningful cost shift for enterprise teams running AI agents and automated workflows at scale.

The headline change is pricing: cache reads for Fable 5.1 drop to $0.25 per million tokens, down from $1.00 for Fable 5. That is a 75% reduction on the cost that matters most when AI agents are running repeated tasks over the same documents and context. Base input and output pricing stays unchanged at $10 and $50 per million tokens respectively.

What Actually Changed

For typical enterprise workloads, Anthropic estimates Fable 5.1 will cost about 25% less than Fable 5. For agentic workloads specifically, those savings jump to around 45%. That gap matters because agentic workflows by design make far more repeated API calls with overlapping context, which means the cache hit rate is higher and the per-call cost reduction compounds.

On performance, Fable 5.1 leads Fable 5 and Opus 5 across every benchmark in Anthropic’s launch comparison. It beats OpenAI’s GPT-5.6 Sol on every row where the two are measured head to head. The sharpest improvement shows on Terminal-Bench-Science, where Fable 5.1 scores 52.6 versus Fable 5’s 24.7 — more than double. That benchmark measures the model’s ability to execute complex multi-step scientific reasoning autonomously in a terminal environment, which translates to more reliable agentic task completion in production.

Fable 5.1 is described as achieving similar or better results than Fable 5 at low and medium effort tiers, with substantially higher ceiling performance when higher effort is used. For businesses running long-horizon tasks such as data analysis pipelines, document review, or customer workflow automation, the improved top-end capability means fewer failures on complex runs.

Where Fable 5.1 Is Available

Fable 5.1 is generally available on the Claude API as claude-fable-5-1, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Multi-cloud availability means enterprise teams already deployed on AWS or Azure do not need to change their infrastructure to access the updated model.

Claude Mythos 5.1, the restricted variant, is available through Anthropic’s controlled-access programs for vetted cybersecurity and life sciences organizations. This is the version that includes capabilities Anthropic keeps behind additional safeguards for sensitive use cases.

What This Means for Business

If your team is using Anthropic’s models for any kind of recurring AI work, the cache pricing change is directly actionable right now.

The biggest beneficiaries are businesses running agentic workflows where the same system prompts, reference documents, or data schemas are passed repeatedly across many calls. Customer support agents, compliance checking tools, report generation pipelines, and similar production use cases all fit this pattern. A 45% cost reduction on those workloads is not a marginal improvement, it is a meaningful shift in the economics of running AI at scale.

For teams evaluating which foundation model to build on, Fable 5.1’s benchmark performance pulling ahead of GPT-5.6 Sol on measured tasks matters for vendor decisions. The competitive balance between Anthropic and OpenAI continues to shift with each release cycle, and teams committing infrastructure to one model family are making a longer-term bet.

The performance gains on science and reasoning benchmarks also signal where Anthropic is investing — autonomous multi-step execution rather than just response quality. That direction aligns with where enterprise AI value is actually being created: not generating better text, but completing more reliable chains of work.

For businesses already using AI agents or considering Omni-style deployments, this release is a reason to re-run your cost projections and benchmark your current workflows against Fable 5.1 before your next infrastructure planning cycle.