Anthropic Launches Claude Fable 5, Its Most Capable Public AI Model Yet

Anthropic rolled out Claude Fable 5 on June 9, 2026, describing it as a Mythos-class model made safe for broad release.

Credit: Anthropic

The company positions Fable 5 as its strongest model available to the public so far, claiming it leads on nearly every benchmark tested, particularly in software engineering, knowledge work, vision, and scientific research.

Fable 5 shares capabilities with the more restricted Claude Mythos 5 but includes tighter guardrails. Queries touching on risky areas, such as advanced cybersecurity or biological research, route to Claude Opus 4.8 instead.

Anthropic tuned these safeguards conservatively; they trigger in fewer than 5% of sessions on average but can block some harmless requests.

The company says it plans to refine them over time.

Pricing and Access

The model carries a price tag of $10 per million input tokens and $50 per million output tokens on the API, less than half the cost of the earlier Mythos Preview.

It offers a 1 million token context window and up to 128,000 output tokens, with a knowledge cutoff in January 2026.

Fable 5 became available immediately across Claude.ai, the API, and related tools for Pro, Max, Team, and certain Enterprise users at no extra charge until June 22.

After that date, usage shifts to paid credits, though Anthropic indicated it may restore broader inclusion once capacity stabilizes.

Claude Mythos 5, the version without the public safety classifiers, went to a limited set of Glasswing partners focused on cyber defense, with plans to expand access to select biology researchers soon.

Performance Claims

Anthropic highlighted several early test results. Stripe reported that Fable 5 handled a massive Ruby codebase migration in a single day, work that would have taken a team months.

On Cognition’s FrontierCode evaluation, it posted the top score among frontier models for producing high-quality production code.

Credit: Anthropic

In knowledge work, it scored highest on Hebbia’s Finance Benchmark for senior-level reasoning.

Vision tasks showed strong gains, including rebuilding web apps from screenshots and completing Pokémon FireRed using only raw game images with minimal tools.

Developer Simon Willison spent hours testing the model on real projects, including upgrading a MicroPython WASM library to full CPython support and adding advanced tool-calling features to his Datasette Agent and LLM libraries.

He described it as feeling “big” in knowledge depth and capable of turning stretch goals into completed work that felt like days of progress in a single session.

Safety and Context

The release follows Anthropic’s earlier limited preview of Mythos-class capabilities to cyber defenders via Project Glasswing.

The company cited risks of misuse in cybersecurity and biology as reasons for the phased approach. Fable 5’s safeguards aim to prevent providing actionable assistance to malicious actors while preserving utility for legitimate work.

Dianne Penn, Anthropic’s head of product management, research and labs, said the goal was to deliver this level of intelligence safely:

“We wanted to be able to provide this level of intelligence for general users in a safe manner.”

Analysts and users noted the model’s strength in long-running, autonomous tasks, where its performance gap over previous Claude versions widens.

Some observers pointed to capacity constraints driving the temporary free-access window, with high demand expected.

Fable 5 arrives amid broader industry moves toward more powerful models, with Anthropic betting that robust classifiers and selective access can balance capability and risk.

The company published a system card detailing safety evaluations.

Initial hands-on reports suggest the model delivers noticeable jumps in complex coding, research, and agentic workflows, though its higher cost and occasional guardrail triggers will shape how teams adopt it. Access is live now for eligible users.