Keep pulling the thread on Claude Fable 5.
Anthropic has officially launched Fable-5, the first model in its Mythos-class series.
Anthropic is deploying Mythos 5 to Project Glasswing partners in collaboration with the US government as an upgrade to Claude Mythos Preview.
On the ExploitBench cybersecurity benchmark, Anthropic's Mythos-5 and Fable-5 scored 78%, compared to GPT-5.5's score of 34%.
On the HealthBench benchmark, Anthropic's Mythos-5 and Fable-5 scored 66%, compared to GPT-5.5's 51.8%.
On the legal agent benchmark, Anthropic's Mythos-5 and Fable-5 scored 13.3%, while GPT-5.5 scored 2.1%.
On GDPval's test of economically valuable tasks, Mythos/Fable 5 scored 1932, surpassing Opus 4.8 (1890) and GPT-5.5 (1769).
On the SuiBench Pro coding benchmark, Mythos and Fable 5 scored 80.3%, significantly higher than Opus 4.8 (69.2%) and GPT-5.5 (58.6%).
On the Frontier Code benchmark, Mythos and Fable 5 scored 29.3%, more than double the score of Opus 4.8 (13.4%) and significantly higher than GPT-5.5 (5.7%).
On Every's senior engineer benchmark, Fable 5 scored 91 out of 100, compared to 63 for Opus 4.8 and 62% for GPT-5.5.
On Cursor's CursorBench, Fable 5 scored 72.9%, which is 8 percentage points above the previous best model.
API costs for Anthropic's Fable-5 are set at $10 per million input tokens and $50 per million output tokens.
Anthropic will remove Fable-5 from its subscription plans on June 23rd, after which access will be on a pay-per-usage basis.