Claude Opus 5 Released: Anthropic's Faster and More Efficient AI Model
Published July 24, 2026
Anthropic has released Claude Opus 5. The new model targets coding, knowledge work and multi-step tasks — according to the company, it nearly matches the stronger Claude Fable 5 at only about half the cost per task.
Compared to its predecessor Claude Opus 4.8, Anthropic promises more performance at unchanged API prices. Opus 5 also becomes the default for Claude Max users and the most capable model in the Claude Pro subscription.
Big gains in coding and problem solving
The largest improvements are in tasks requiring autonomous planning, programming and tool use. On the Frontier-Bench coding benchmark, Opus 5 scores 43.3 percent — up from just 21.1 percent for Opus 4.8. New "effort" settings let users trade speed and cost against more thorough reasoning.
As usual, vendor benchmarks should be treated with caution — most results come from Anthropic itself or selected partners, and independent testing is still pending.
Better self-verification for AI agents
A key upgrade, says Anthropic, is stronger self-control: Opus 5 checks its own results more often, detects errors and adjusts its approach. In one test, the model had to build a 3D part from a drawing — unable to read the image directly, it developed its own method to extract the geometry. In another, it traced a bug in an open-source package manager back to its root cause, not just the visible symptom.
Such capabilities matter most for AI agents that work through longer tasks, modify code and operate software on their own.
Enterprise results and safety
Cursor, Cognition, Zapier, Box, Stripe and JetBrains tested the model early. Zapier reports Opus 5 ran a full analysis of at-risk customer accounts and notified the responsible staff; Box cites improvements in data analysis and due diligence. Anthropic also calls Opus 5 its best-aligned model yet, while penetration testing and exploit development remain restricted by safety filters.
Pricing and availability
Claude Opus 5 is available on the Claude platforms and via the API as claude-opus-5, at the same prices as Opus 4.8:
- $5 per million input tokens
- $25 per million output tokens
An optional Fast Mode runs about 2.5x faster at twice the price.
Bottom line
Claude Opus 5 is built for longer, more complex work: the biggest leap is not just better answers, but more reliable behavior — planning ahead, catching its own mistakes and verifying results before handing them over. It targets developers and companies that want capable AI agents without paying flagship prices for every task. Independent real-world tests will show how large the improvement really is.
