Subscribe The AI Brief

Tuesday, September 1, 2026

Today’s Edition

AI Intel Report

MARKETS —

Frontier Models

Anthropic Releases Claude Fable 5.1 and Mythos 5.1 With Record Coding Scores

The models deliver major benchmark gains in coding tasks along with cost reductions and differentiated safeguards, with one version open to the public and the other restricted to vetted organizations.

By Marcus Vance 3 MIN READ
Inside a spacious open-plan technology research office at dusk with large floor-to-ceiling windows showing city lights outside, several anonymous figures wearing casual business attire sit at long shared wooden desks equipped with multiple flat-panel monitors, mechanical keyboards, and external hard drives connected by thick black cables. The figures are viewed from behind or in three-quarter profile with faces obscured by angles or hair, each focused intently on programming workflows across their screens that display only abstract blurred patterns of syntax without any legible characters. On one side of the room a cluster of three figures collaborates around a central laptop while pointing at shared code repositories and test result dashboards, representing the public-access version of advanced coding models, while on the opposite side of the same continuous space a separate pair of figures works in a cordoned alcove behind a frosted glass partition symbolizing restricted access for vetted organizations. Scattered across the desks are physical objects including open notebooks with blank pages, USB drives labeled only with color stickers, stacks of technical reference books with closed covers, and small potted plants. In the background rows of server racks with blinking status lights line the far wall next to a whiteboard covered by a large sheet of paper to prevent any markings from being visible. The entire environment features neutral gray and blue tones, modern ergonomic chairs, overhead pendant lighting, and visible network cables running along the floor in neat conduits. Additional details include a water cooler in the corner, a coat rack with jackets, a coffee mug on each desk, a small potted succulent near the monitors, and subtle reflections on the polished concrete floor. The scene captures the atmosphere of high-stakes software development and benchmark evaluation through the arrangement of hardware, the concentration of the anonymous workers, the division between open and restricted workspace zones, and the presence of terminals and data storage devices that evoke performance testing in coding benchmarks without any visible text, logos, numbers, or user interface elements anywhere in the frame.
Illustration: AI Intel Report

Claude Fable 5.1 and Claude Mythos 5.1 are versions of the same underlying model from Anthropic that provide advanced capabilities for coding and knowledge work with different safeguard configurations.

Anthropic released Claude Fable 5.1 on September 1, 2026. The model features a 1 million token context window and a maximum output of 128 thousand tokens. Pricing stands at 10 dollars per million tokens for input and 50 dollars per million tokens for output.

Claude Fable 5.1 is generally available to the public. Claude Mythos 5.1 is available only through trusted access programs for vetted organizations in cybersecurity and life sciences.

What are the key specifications of Claude Fable 5.1?

Claude Fable 5.1 defaults to high effort in Claude Code and medium in other interfaces. Cache reads for Claude Fable 5.1 cost a quarter of the prior price.

How do the benchmark scores compare to previous models?

Claude Fable 5.1 scores 52.6 percent on Terminal-Bench-Science 0.1 compared to 24.7 percent for the prior model. On Terminal-Bench 4.0 Claude Fable 5.1 achieves 55.8 percent accuracy versus 42.0 percent previously.

Claude Mythos 5.1 reaches 60.9 percent on coding benchmarks. These scores position the models as leaders in terminal based coding evaluations.

Benchmark performance and cost metrics for the new Claude models
ModelTerminal-Bench-Science 0.1Terminal-Bench 4.0Typical Cost Reduction
Claude Fable 5.152.6%55.8%~25%
Claude Mythos 5.1Not specified60.9%Not specified

What cost reductions are associated with Claude Fable 5.1?

Cache reads for Claude Fable 5.1 cost a quarter of the prior price. This change results in approximately 25 percent lower typical workload costs compared to Fable 5.

How do the safeguards differ between the models?

In cybersecurity the newest safeguards block 60 percent fewer false positives than before. The biology safeguards intervene on benign requests 85 percent less often than the ones launched with Fable 5.

What expert reactions have emerged regarding the models?

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. While prior models became hard to follow the longer they worked, Fable 5.1 remains readable over long, multi-step tasks.Craig Falls, Head of Quantitative Research

What are the market implications of these releases?

Claude Mythos 5.1 targets organizations in cybersecurity and life sciences through trusted access programs. Project Glasswing participants can access the capabilities of Claude Mythos 5.1.

The cost reductions make the models more viable for extended use in coding tasks across multiple sectors.

What comes next in the development of these models?

Anthropic continues to focus on refining safeguards for safe frontier deployment in sensitive fields.

  1. Assess eligibility for trusted access programs for Mythos 5.1.
  2. Evaluate integration with existing coding workflows using the new context window.
  3. Monitor updates on further benchmark improvements from Anthropic.

Frequently asked

What is the difference between Claude Fable 5.1 and Claude Mythos 5.1?

Claude Fable 5.1 is the publicly available version with standard safeguards. Claude Mythos 5.1 is the same underlying model with enhanced safeguards and is limited to trusted access programs for vetted organizations in cybersecurity and life sciences.

Sources

  1. Anthropic — Benchmark scores and cost reductions for Claude Fable 5.1 and Mythos 5.1
  2. Anthropic — Details on safeguards for Mythos 5.1 and comparison to Fable 5.1
  3. Anthropic — Release date, specifications, and availability of Claude Fable 5.1 and Mythos 5.1
  4. Anthropic — 75% cheaper (quarter of the cost) — Cache read pricing reduction for Claude Fable 5.1