Reviewed: Claude Opus 5 lands at half Fable 5's price and takes Frontier-Bench outright

Anthropic shipped Claude Opus 5 on July 24, its fourth model in under two months, and the pitch is unusually blunt: Opus 4.8 pricing ($5 in, $25 out per million tokens), Frontier-Bench v0.1 taken outright, and a marketing line, "Opus 5 comes close to Fable 5 intelligence at Opus speed and cost", that reads as a direct assault on its own flagship's token economics.

The cadence is worth pausing on. Opus 4.8 landed May 28, per TechCrunch's Russell Brandom. Mythos 5, Fable 5, and Sonnet 5 all shipped in June. Fable 5 was pulled on June 12 after Amazon researchers reported bypassing its safeguards, then re-released June 30 with hardened security. Haiku alone still sits on the 4-series. This isn't a release schedule so much as an admission that the previous six weeks didn't quite land, and Opus 5 is being positioned as the version that consolidates the lineup.

On the numbers, Anthropic says Opus 5 "more than doubles Opus 4.8's performance" on Frontier-Bench v0.1, and claims new state-of-the-art on GDPval-AA. A FreeCAD-based demonstration reports no competing model solving the task in five attempts under the same setup. Cursor's team places Opus 5 "just under Fable 5" on CursorBench. Zapier says it tops AutomationBench with a 100% pass on a churn-prevention workflow that had previously failed. Anthropic concedes Mythos 5 remains ahead on cybersecurity and still recommends Fable 5 for multi-day autonomous work.

The two structural features matter more than any single benchmark. The five-step effort dial (Fortune describes low/medium/high stops) exposes the cost/capability trade-off as an API parameter, which is Anthropic's response to what Fortune characterized as customer complaints about Fable 5's "burn rate." Anthropic's own launch post shows performance curves plotted against effort as the independent variable. It's an admission that flat pricing was hiding a knob customers wanted to turn themselves.

Automatic Fallbacks, in beta, routes requests to a less capable model when Opus 5's safety classifier engages. Anthropic expects classifiers to fire 85% less often than on Fable 5, which Brandom notes reflects "the lighter touch given to the less capable model." Refusals stop being terminal. For enterprise buyers building on the API, that's the release's real unlock, and Bloomberg's Shirin Ghaffary frames the whole launch as Anthropic's answer to cost-conscious buyers and rising Chinese competition.

Opus 5 becomes the default on Claude Max and the strongest option on Claude Pro. The through-line across every design decision, pricing held flat, refusals routed rather than returned, effort exposed as a dial, is a company that has decided the constraint on adoption isn't capability. It's the bill and the friction.

Sources

• https://www.anthropic.com/news/claude-opus-5
• https://techcrunch.com/2026/07/24/anthropic-launches-opus-5/
• https://www.bloomberg.com/news/articles/2026-07-24/anthropic-unveils-more-cost-efficient-model-for-everyday-tasks
• https://www.cnbc.com/2026/07/24/anthropic-claude-opus-5-ai-fable-5-cost.html
• https://fortune.com/2026/07/24/anthropic-debuts-claude-opus-5-with-feature-that-lets-users-toggle-between-cost-and-capability/