Anthropic released Claude Opus 5 Friday with pricing at half that of Fable (the previous flagship), claiming near-parity performance. On Epoch's Capabilities Index (ECI), Opus 5 scores 159 vs. Fable 5's 161—technically below but within noise. On software engineering (SWE-ECI), Opus 5 matches Fable 5 at 161. Artificial Analysis ranks Opus 5 as the leader on its Briefcase agentic knowledge benchmark, beating Fable by ~150 Elo while reducing cost-per-task by 20%.
Early user reports emphasize Opus 5's coding and browser-automation performance, with practitioners describing it as a clear winner on math and tool use versus Fable despite narrower benchmark margins. The tight official benchmark gaps reflect what Anthropic acknowledges: difficulty in measuring 'big model smell' and qualitative robustness that evals don't capture. Nous Research is offering a 20% discount on Opus 5 access through its portal.
For architects: this is a price-to-performance inflection. Opus 5 delivers frontier-tier agentic capabilities (browser control, multi-step reasoning, tool chaining) at a cost structure closer to smaller models. For production inference on coding, workflows and knowledge tasks, the 50% cost reduction makes frontier models economically viable at scale without sacrificing capability—effectively narrowing the gap between Anthropic and faster, cheaper alternatives like Grok Imagine and DeepSeek.