Claude Opus 5 outperforms Fable 5 on most benchmarks

Anthropic released Claude Opus 5 on July 24 and priced it at $5 per million input tokens. The company set Opus 5 as the default on Claude Max and the top option on Claude Pro, replacing Claude Fable 5 for many subscribers.

On Frontier-Bench v0.1, which tests whether AI coding agents can complete end-to-end software engineering tasks, Opus 5 scored 43.3%, compared with 33.7% for Fable 5 and 34.4% for OpenAI’s GPT-5.6 Sol. On ARC-AGI-3, a benchmark of novel problem-solving, Opus 5 scored 30.2% while GPT-5.6 Sol scored 7.8%; Fable 5 was not entered on that test. On GDPval-AA v2, a knowledge-work evaluation using Elo-style ratings, Opus 5 reached 1,861, versus Fable 5 at 1,747 and GPT-5.6 Sol at 1,736.

Companies that ran operational tests reported practical improvements. Zapier reported that Opus 5 completed a full churn-prevention workflow end to end — flagging at-risk accounts, alerting owners and producing a summary — and that the model “hit 100%.” Ultima Genomics wrote that Opus 5 “behaves more like a careful scientist than any model we’ve run,” noting the model selects statistical tests, cross-checks results and stays on task during long analyses. Lovable, a development platform, noted roughly a 22% gain on its hardest agentic coding tasks versus Opus 4.7 and reported less variance between runs.

Claude Fable 5 had a limited rollout earlier this summer. The model launched on June 9, was pulled three days later after the U.S. government issued an emergency export control order citing a jailbreak vulnerability, and returned on June 30 under a credits-only model removed from standard subscription plans. With Opus 5 available via API and subscriptions, Anthropic has made a lower-cost option accessible to more paying users.

Opus 5 is available via API at $5 per million input tokens and $25 per million output tokens. A Fast mode, marketed at about 2.5 times the default speed, is offered at double the base price: $10 per million input and $50 per million output. Tokens are the units of text that models process in input and output.

The Opus 5 release follows a recent launch by Moonshot AI of Kimi K3, a 2.8-trillion-parameter open-weight model. Independent benchmarks place Kimi K3 behind leading closed models on many tasks while outperforming them in specific areas.

Anthropic described Opus as its heavy-duty option within a four-tier lineup that also includes Haiku for low-cost, Sonnet for mid-range needs and the higher-restriction Mythos tier, which includes Fable 5 for public use and a more restricted Mythos variant available to vetted researchers and critical infrastructure operators.