Fable took the board

List price did not move on input and output. Anthropic: $10 per million input tokens, $50 per million output tokens, unchanged from Fable 5. Cache reads now $0.25 per million, down from $1.00. That is the 75 percent cut. Five-minute cache writes stay $12.50 per million; one-hour writes $20. Batch API is $5 input and $25 output. Anthropic’s platform table is the source. A VentureBeat comparison table on 1 September listed Claude Fable 5.1 / Claude Mythos 5.1 as $10.00 / $50.00 / $60.00. That third figure is input plus output, not a third rate. The official pair remains $10 and $50. I'm not inventing a $60 output price. It is not on Anthropic’s page.

Artificial Analysis supported Anthropic with pre-release evaluation. At max effort, Fable 5.1 scores 66 on the Intelligence Index. At xhigh effort it scores 65, at $2.72 per Index task against $3.76 at max. Opus 5 at max is 63, at $2.34 per task. Fable 5 at max was 62. GPT-5.6 Sol at max, 61. Grok 4.6 at high, 61. Those are Artificial Analysis’s numbers, 1 September, not Anthropic’s marketing table. The lab evaluated with Anthropic’s default server-side fallback, which routes safety-flagged requests to Claude Opus 4.8 or Claude Opus 5; fallback served about 4 percent of output tokens across the Index. On Humanity’s Last Exam, Artificial Analysis has Fable 5.1 at 59.1 percent, ahead of Fable 5’s 55.5 percent. Terminal-Bench v2.1: 91.4 percent. SciCode: 62.0 percent. GDPval-AA v2: 1,853 Elo, with the lead over Opus 5’s 1,824 inside the confidence interval. AA-Briefcase: 1,694 against Opus 5’s 1,685, effectively tied.

Anthropic’s own table is a different bench, and it is vendor-reported. Terminal-Bench-Science 0.1: Fable 5.1 at 52.6 percent, Fable 5 at 24.7, Opus 5 at 29.0. Terminal-Bench 4.0: Fable 5.1 at 55.8 percent; Mythos 5.1 at 60.9 percent under more permissive cyber safeguards. CursorBench 3.2.0: 73.4 percent. Attribute the table. It is Anthropic’s. Artificial Analysis is the independent Index.

Context window is 1 million tokens. Max output 128,000 tokens. Adaptive thinking is always on. Default effort on the Claude API is high. Knowledge cutoff June 2026. Platforms on Anthropic’s model page: Claude API as claude-fable-5-1; Amazon Bedrock as anthropic.claude-fable-5-1; Google Cloud claude-fable-5-1; Microsoft Foundry claude-fable-5-1; Claude Platform on AWS the same id. Anthropic’s announcement also names Amazon Web Services, Google Cloud, and Microsoft Azure. Released 1 September 2026. Retirement not sooner than 1 September 2027.

Anthropic’s cost claim is not the Index task price. “Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads, wherever usage is billed by token.” For highly agentic work, “up to approximately 45%.” The cut is cache reads. Artificial Analysis, measuring its own Index tasks, still has Fable 5.1 at max costing 20 percent more per task than Fable 5 at max, because it uses about 1.7 times the output tokens. The cache cut saves about $1.40 per Index task; without it the max-effort task would have been about $5.16. Both sentences can sit on the same page. One is Anthropic on typical billed workloads. One is Artificial Analysis on Index tasks. They are not the same meter.

Mythos 5.1 is not a second brain. It is the same model with a different lock. Anthropic’s system card, 1 September: two configurations, identical weights. Fable 5.1 for general use, with additional safeguards that block certain tasks in high-risk dual-use domains such as biology and cybersecurity. Mythos 5.1 relaxes some of those domain-specific safeguards for vetted individuals and organisations. Direct access is limited to trusted access programmes. Anthropic names two: the Cyber Verification Program, which currently covers certain Opus- and Sonnet-class models with reduced cyber safeguards for defensive security work and, “in the near future,” Mythos-class models; and the Life Sciences Verification Program, designed so life-sciences professionals can use Mythos 5.1 with safeguards built for professional research. Anthropic says Mythos 5.1 is currently available to a set of US organisations, with expansion coordinated with the US government. Claude Security, which scans codebases and suggests patches for human review, is powered by Mythos 5.1 and is listed for Claude Enterprise customers.

Project Glasswing is the named defensive umbrella. Anthropic’s Claude docs: Mythos 5.1 “offers the same capabilities by invitation only, as part of Project Glasswing.” Contact your Anthropic, AWS, or Google Cloud account team. That is the access sentence. It is not a download link. Fable 5.1 is the public product. Mythos is the vetted door.

Anthropic’s launch line is a claim, and it stays in quotes. “They’re the world’s most advanced models for coding and knowledge work—and their research capabilities offer an early glimpse of how AI models will contribute to scientific progress.” The Index is the independent score. 66. Opus 5 is 63. The price that changed is cache reads, $0.25, down 75 percent. Input $10. Output $50. API id claude-fable-5-1. AWS, Google Cloud, Azure / Microsoft Foundry. 1 September 2026. That is the story. A model as a product. A bench as news. Not a hymn.

Jane Street’s Craig Falls, on Anthropic’s page: “In internal benchmarks, Claude Fable 5.1 solves more of our coding problems than Fable 5 or Opus 5, and achieves state of the art on trading intuition.” Cursor, on the same page, called it the most capable model it had run on CursorBench 3.2 at 73.4 percent max effort. Partner notes. Not my stopwatch.

What you can stand behind today: two names, one weight file, two safeguard settings, a public API id, a 1-million-token window, a 75 percent cache-read cut, $10 / $50 on the official card, and an independent Index lead at 66 against Opus 5 at 63. Mythos is the same model for vetted cyberdefence and life-sciences work through Project Glasswing and the verification programmes. The $60 in one VentureBeat table is a sum. It is not a third Anthropic rate. Name the products. Name the board. Leave the rest.