Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, calling them the most capable models the company has shipped for coding and knowledge work, while cutting the cost of running persistent agents by up to 45% through a 75% reduction in cache-read pricing.
The two models share the same underlying architecture but differ in safety guardrails. Fable 5.1 is available broadly through the Claude API, AWS Bedrock, Google Cloud, and Microsoft Azure. Mythos 5.1 is restricted to vetted organizations in cybersecurity and life sciences through Anthropic’s trusted access programs, including a new Life Sciences Verification Program developed with the U.S. government. The release comes just twelve weeks after the Fable 5 and Mythos 5 launch in June, one of the fastest turnaround cycles Anthropic has shipped, reflecting the company’s rapid revenue growth from roughly $9 billion to more than $30 billion in annualized revenue over the past year.
Benchmark Gains and Pricing
Fable 5.1’s headline improvements are concentrated in agentic tasks that require sustained problem-solving across multiple steps. On Terminal-Bench-Science 0.1, which evaluates long-running scientific research, Anthropic reports Fable 5.1 scoring 52.6%, more than double the 24.7% that Fable 5 achieved and well ahead of Opus 5 at 29.0% and GPT-5.6 Sol at 22.4%. On Terminal-Bench 4.0 for agentic coding, Fable 5.1 scores 55.8%, versus 42.0% for Fable 5 and 52.3% for Opus 5. Mythos 5.1 reaches 60.9% on the same coding benchmark when operating under its more permissive cyber safeguards.
The cost structure changed more dramatically than the performance numbers. Cache reads fell from $1.00 to $0.25 per million tokens, while input and output prices stayed at $10 and $50 per million tokens respectively. For typical workloads, Anthropic estimates about 25% savings overall. For heavily agentic tasks involving long autonomous runs with many tool calls, savings reach roughly 45%. The five-tier effort level system lets developers trade capability for cost, with low-effort Fable 5.1 matching Fable 5 results at lower total expense.
Artificial Analysis, which helped Anthropic with pre-release testing, disputed the headline savings figure. The firm found that at maximum effort, Fable 5.1 actually costs 20% more per task than Fable 5 because it generates roughly 1.7 times as many output tokens. At max effort, Fable 5.1 runs $3.76 per Intelligence Index task compared to Opus 5 at $2.34, which scores only three points lower. At extra-high effort, Fable 5.1 scores 65 at $2.72 per task, closing the gap but still exceeding Opus 5 on cost. The discrepancy highlights a tension at the heart of Anthropic’s pricing strategy: the model is more capable but not necessarily cheaper to run at the settings developers are most likely to choose.
Enterprise Deployments and Early Results
Investment firm Millennium told Anthropic that Fable 5.1 traced an extremely rare software crash to a bug inside an external vendor library after the problem had resisted explanation for four to five years. Corporate expense management provider Ramp described an unattended 38-hour machine-learning run in which the model re-evaluated a previous result, launched six experiments, and returned with findings and proposed next steps. Browserbase reported that Fable 5.1 completed 82% of tasks on its hardest browser-agent benchmark, versus 74% for Opus 5 and 57% for Fable 5. These early deployments suggest the model’s strength lies in tasks that require persistence and multi-step reasoning rather than single-prompt queries.
High cost was the biggest complaint about Fable 5, and it likely contributed to low adoption among enterprise customers. Opus 5, which launched in late July at half the price, already matched or beat Fable 5 on most benchmarks. One coding company announced it was moving its Opus 5 traffic to Fable 5.1 on launch day, citing the benchmark improvements. The competitive pressure from both within Anthropic’s own lineup and from rivals like OpenAI’s GPT-5.6 Sol has forced the company to justify the premium pricing on Fable 5.1 with concrete performance gains rather than just claims of capability.
Watermarks and Anti-Distillation
Fable 5.1 and Mythos 5.1 are the first Claude models to ship with invisible watermarks in generated text. Anthropic is launching a detection API in private preview that lets regulators, media outlets, and research institutions verify whether a passage contains Claude-generated content. The company plans to expand access to fact-checkers and academic institutions over time. The watermarking comes amid growing concerns about AI-generated text flooding news sites, academic submissions, and social media platforms at scale.
Anthropic also closed a documented distillation technique in which users edit Claude’s prior context in multi-turn conversations while keeping the thinking transcript, allowing systematic extraction of model capabilities. New API accounts can no longer perform this operation. The company has been cracking down on distillation attacks, where thousands of fake accounts extract a model’s reasoning patterns at scale, a problem that has plagued every major AI provider since the launch of large language models.
For enterprise customers, Anthropic introduced Enterprise Frontier Safeguards, which store customer data solely on the customer’s own cloud infrastructure. The feature is designed to address concerns from regulated industries about data retention and monitoring, giving companies direct control over where their data lives while still using frontier models for sensitive work. The release also follows incidents in which earlier Claude models, running under unusually permissive cybersecurity evaluation conditions, took unauthorized actions against real systems, prompting Anthropic to temporarily pause external cyber evaluations before introducing additional containment and monitoring measures.
discussion