Key Notes
- Anthropic released Claude Fable 5.1 on September 1 as its most capable generally available model for coding, knowledge work and long-running problem solving.
- The model keeps Fable 5's $10 per million input and $50 per million output token prices, but cache reads are 75% cheaper, reducing estimated costs by about 25% for typical workloads and up to 45% for highly agentic work.
- Anthropic reports major gains over Fable 5 across coding, scientific research, knowledge work, computer use and long-context tasks.
Anthropic has released Claude Fable 5.1, an upgraded version of its flagship model for coding, research and complex knowledge work. The company says the new model sets higher performance marks across demanding evaluations while also lowering the effective cost of long-running workloads.
Fable 5.1 is available immediately to Pro, Max, Team and Enterprise users, as well as through the Claude API and major cloud platforms. Anthropic also launched Claude Mythos 5.1, the same underlying model with stronger safeguards designed for vetted cybersecurity and life-sciences research partners.
The headline token prices have not changed. Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, matching Fable 5. The major pricing change is cache reads, which now cost $0.25 per million tokens, 75% less than with Fable 5. Anthropic estimates that this cuts the practical cost of typical workloads by about 25% and highly agentic workloads by up to 45%.
Fable 5.1 Pushes Performance Higher
Anthropic positions Fable 5.1 as its most capable generally available model, with improvements concentrated in long-running agentic coding, multistep research, documents, spreadsheets, presentations, vision, long-context reasoning and computer use.
On Terminal-Bench-Science 0.1, which measures agentic scientific research, Fable 5.1 scored 52.6%, compared with 24.7% for Fable 5 and 29.0% for Opus 5 in Anthropic’s evaluation. On Terminal-Bench 4.0 for agentic coding, the new model reached 55.8%, versus 42.0% for Fable 5 and 52.3% for Opus 5.
Fable 5.1 also scored 1,853 on GDPval-AA v2, a benchmark for real-world knowledge work. That compares with 1,723 for Fable 5 and 1,824 for Opus 5. On OSWorld 2.0, Anthropic reports 77.9% on its partial metric and 41.7% on the strict metric, both above the corresponding Fable 5 results.
These are Anthropic’s own benchmark results, and the company notes that its production safeguards were enabled during evaluation. Some safeguards can route particular cybersecurity and biology tasks to other models, which can affect benchmark results.
The company also says Fable 5.1 can achieve similar or better results than Fable 5 at lower effort levels. That matters for cost because users can trade some reasoning depth for lower token consumption without necessarily giving up the performance of the previous flagship.
Designed for Longer AI Agent Workflows
Fable 5.1 keeps the 1-million-token context window and 128,000-token maximum output of Fable 5. Adaptive thinking is always enabled, while an effort setting allows developers to control how deeply the model reasons.
Anthropic has also added per-message effort controls in beta, allowing the reasoning level to change during a conversation. A developer can use higher effort for a difficult step and reduce it for routine work without restarting the session.
The model is aimed particularly at agentic AI workflows that can run for hours. Anthropic highlights large codebase refactors, debugging, code review, research, spreadsheet construction, slide creation and computer-use tasks as areas where Fable 5.1 is designed to work more reliably over extended sessions.
That puts Fable 5.1 directly into the growing market for AI systems that do more than answer individual prompts. Anthropic’s own Claude model guide describes Fable as the top end of its model family, with Opus, Sonnet and Haiku occupying lower points on the performance and cost curve.
Fable 5.1 also changes how developers manage conversations. Earlier Claude models cannot read Fable 5.1’s thinking blocks, while Fable 5.1 can read thinking blocks produced by earlier models. Anthropic recommends treating conversation history as append-only and avoiding edits to earlier turns that would invalidate preserved thinking blocks.
Cheaper Cache Makes Frontier AI More Practical
The biggest economic change is the reduction in cache-read pricing. Long-running agents repeatedly reference previously processed context, so the cost of reading that cached material can become a significant part of the total bill.
Fable 5.1 reduces that rate to one-quarter of the Fable 5 level. Anthropic says the change should lower costs by about 25% for typical workloads and by as much as 45% for highly agentic applications that repeatedly reuse large context windows.
That makes the model more competitive for software teams that run autonomous coding agents for extended periods. A developer can keep a large repository, task history and tool context available without paying the full input rate every time the agent revisits information it has already processed.
The pricing strategy also creates a clearer distinction between headline token prices and the actual economics of agentic AI. The base input and output rates remain unchanged, but the workload patterns associated with autonomous agents can produce substantially lower effective costs under Fable 5.1.
Anthropic has simultaneously expanded the model’s availability. Fable 5.1 is offered through the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry, giving enterprises several routes to deploy it without changing the underlying model.
The release comes less than three months after Anthropic introduced Fable 5 for general use. AIstify reported at the time that Fable 5 became the company’s most capable model available to the public.
With Fable 5.1, Anthropic is trying to make the flagship tier more useful for work that lasts hours or days rather than individual interactions. The combination of higher benchmark performance, a million-token context window and sharply lower cache costs is aimed at making frontier-level AI agents practical for a wider range of professional workloads.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
