Anthropic’s Claude Opus 5 keeps prices flat while matching or beating Fable 5 on key benchmarks

智东西 · wechat · 2026-07-25

Anthropic’s Claude Opus 5 is presented as a major update that narrows the gap to Fable 5 while keeping pricing flat.

Pricing stays at $5 / $25 per million tokens for input/output, the same as Opus 4.8.
In coding and knowledge-work benchmarks such as Frontier-Bench and GDPval-AA, Opus 5 reaches SOTA or near-SOTA results; on CursorBench 3.2 it comes within 0.5% of Fable 5’s peak at max effort while costing about half as much per task.
The model is also described as stronger in scientific work, especially chemistry and protein-related tasks, and more capable at visual output generation.
Anthropic says it improves self-checking, iterative problem solving, and safety, with lower deception rates and stronger constitutional alignment than previous Claude models.
New test-phase features include mid-conversation tool-prompt changes without cache invalidation and automatic fallback handling for classifier-flagged requests.

More from Models