Anthropic releases Claude Opus 5, closing the gap on its own flagship at half the price

Anthropic released Claude Opus 5 on July 24, 2026, the newest version of its mid-tier flagship reasoning model, available immediately at the same price as its predecessor.

What's new

According to Anthropic's announcement, Opus 5 is "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price." The company says the new model is state of the art on Frontier-Bench and GDPval-AA, two evaluations used to gauge real-world coding and knowledge-work performance, and lands within 0.5% of Fable 5's peak score on CursorBench 3.2 while costing half as much to run.

On software engineering work specifically, Anthropic says Opus 5 "excels on valuable software engineering tasks" and "more than doubles Opus 4.8's performance at a lower cost per task." The model also posts a large jump on ARC-AGI 3, a benchmark built around novel problem-solving rather than memorized patterns: "On ARC-AGI 3, an evaluation where the model has to solve novel problems, Opus 5's score is three times as high as the next-best model."

Anthropic also highlights gains in scientific domains, citing improved performance on life sciences, organic chemistry, and protein-related tasks. Pricing holds steady at $5 per million input tokens and $25 per million output tokens — identical to Opus 4.8's rate card, meaning the capability increase comes at no extra cost to existing customers.

Context

Opus 5 replaces Opus 4.8 as Anthropic's mid-tier model, sitting below Fable 5, the company's top-line flagship, in its three-tier lineup alongside Sonnet 5 and Haiku 4.5. The release lands amid a run of other Anthropic announcements this month, including an expanded Managed Agents API with effort levels and lifecycle webhooks, a $200 million Economic Futures Research Fund, and final court approval of its $1.5 billion AI copyright settlement — all part of a broader push to scale both its research and its commercial platform simultaneously.

The comparison Anthropic draws to its own Fable 5 model, rather than to competitors, is notable. Rather than simply improving Opus in isolation, the company is explicitly narrowing the distance between its cheaper and most expensive models, a strategy that puts pressure on the value proposition of paying for the top tier at all.

Why it matters

A model that closes to within half a percentage point of your own flagship, at half the price, is as much a pricing move as a capability one. For developers building production agents and coding tools, Opus 5's positioning — near-frontier reasoning without frontier-tier cost — makes it a more attractive default than reaching for Fable 5, and squeezes competitors whose mid-tier offerings (OpenAI's GPT-5.6 family, Google's Gemini lineup) now have to match both the price and the benchmark numbers. The 3x jump on ARC-AGI 3 in particular suggests Anthropic has made real progress on generalization rather than just benchmark-specific tuning, which is the harder problem to solve and the one most likely to show up in unpredictable real-world agentic tasks.