Anthropic launches Claude Fable 5 at $10, claims top benchmarks

Anthropic released Claude Fable 5 on Tuesday, offering a Mythos-class model for general use priced at $10 per million input tokens and $50 per million output tokens. Developers can call the model via the Claude API using the string claude-fable-5. Fable 5 is available on Pro, Max, Team and Enterprise plans at no extra charge through June 22; starting June 23 the company will require usage credits and says it plans to return the model to standard subscription plans when capacity permits.

The company highlighted benchmark results showing strengths in software engineering, knowledge work, scientific research and long-context tasks. Early users reported concrete gains: Stripe reported compressing more than two months of engineering work into a single day while migrating a 50-million-line Ruby codebase with Fable 5. A FrontierCode comparison showed higher coding accuracy for Fable 5 than Anthropic’s Opus 4.8 and GPT-5.5 as task complexity and cost increased.

In finance and research tests, Fable 5 passed trading analysis checks for factual lookup, conceptual reasoning, root-cause analysis and expected-value calculations, according to IMC. On a senior-level reasoning benchmark from Hebbia, the model scored highest among tested models. A partner working on frontier physics research reported that Fable 5 reached a nearly comparable result to GPT-5.5 after 36 hours while using roughly one-third of the reasoning tokens; GPT-5.5 required about four days to reach the same result.

Anthropic reported vision and long-context gains. The company said Fable 5 completed the video game Pokémon FireRed using only raw screenshots, without maps or navigation tools. On memory-heavy tasks, file-based memory increased Fable 5’s performance on the deck-building game Slay the Spire by a factor of three compared with Opus 4.8 under the same conditions.

Anthropic introduced Claude Mythos 5 alongside Fable 5. Mythos 5 is built on the same underlying model but with fewer cybersecurity restrictions and is being deployed through Project Glasswing in collaboration with the U.S. government as an upgrade to Claude Mythos Preview. Anthropic described Mythos 5 as having the strongest cybersecurity capabilities of any model it has released.

To manage sensitive requests, Fable 5 uses classifiers that reroute queries involving cybersecurity, biology and chemistry, or model distillation to Opus 4.8 instead of refusing them. The company reported that more than 95% of sessions trigger no fallback. A new 30-day data retention policy applies to all Mythos-class model traffic; Anthropic said the data will not be used for model training and will be deleted after 30 days in most cases.

Separately, Anthropic filed confidentially for an initial public offering in recent weeks. OpenAI filed a similar confidential IPO application soon after. Both filings occurred while AI, semiconductor and large-cap tech stocks continued to face selling pressure on Wall Street.