Anthropic's new Claude Opus 5 nearly matches its top-tier Fable 5 model on coding and knowledge work at half the cost — and it ships at the same price as Opus 4.8.
What changed and why it matters
On July 24, 2026, Anthropic introduced Claude Opus 5, and the headline is refreshingly simple: a big jump in capability with no jump in price. Opus 5 comes close to the frontier intelligence of Anthropic's top-tier Fable 5 model, but costs roughly half as much per task. For teams that have been rationing their best model for only the hardest problems, that math changes what you can afford to run every day.
Pricing stays flat at $5 per million input tokens and $25 per million output tokens — identical to its predecessor, Opus 4.8. You get the capability step-up for free. Opus 5 is now the default model on Claude Max and the strongest model available to Claude Pro subscribers.
The benchmarks
Anthropic positions Opus 5 as a new internal state-of-the-art on coding and knowledge-work evaluations. A few of the concrete numbers from the announcement:
- Frontier-Bench v0.1 (coding): surpasses all other models and more than doubles Opus 4.8's score — at lower cost.
- CursorBench 3.2: lands within 0.5% of Fable 5's peak score, at half the cost per task.
- ARC-AGI 3 (reasoning): roughly three times the score of the next-best model.
- Zapier AutomationBench: a pass rate around 1.5× the next-best model.
- OSWorld 2.0 (computer use): surpasses Fable 5's best result at just over a third of the cost.
- Science: +10.2 points over Opus 4.8 on organic chemistry, +7.7 points on protein prediction.
Built to finish the job
The most interesting shift isn't a single benchmark — it's temperament. Anthropic describes Opus 5 as "much stronger at verifying its work and iterating carefully until it succeeds." In practice that means an agent that checks itself and keeps going rather than stopping at the first plausible answer.
The launch examples lean into that: handed a real bug in a popular open-source package manager, Opus 5 traced the root cause and fixed an edge case that the community's own patch had missed. Other cited runs include a FreeCAD reconstruction that required building a custom computer-vision pipeline, and a market-data feed plus test harness assembled for a trading firm in a single session. This is the profile you want for long-running agentic work, where the cost of a confidently-wrong step compounds.
The caveats worth knowing
Opus 5 is not the top model for everything. Anthropic notes it remains behind its Mythos 5 model on cybersecurity-exploitation tasks and in biology research capability — deliberate boundaries on the highest-risk domains. On the safety side, Anthropic reports Opus 5 showed the lowest rates of deceptive behavior in its behavioral audit, and expects its cyber classifiers to intervene far less often than with Fable 5. Full details are in the accompanying system card.
How to get it
Opus 5 is available today across Claude.ai, Claude Code, Claude Cowork, and the API, under the model id claude-opus-5. A "fast mode" runs at about 2.5× the default speed for 2× the base price when latency matters more than cost. It's the fourth model Anthropic has shipped in under two months, after Mythos 5, Fable 5, and Sonnet 5.
Our take
The story of this release is economics, not just capability. When near-frontier quality drops to half the cost — and lands at the same price as last quarter's model — the question stops being "can we afford to use the best model here?" and becomes "why aren't we?" For anyone building coding agents or automating knowledge work, Opus 5 is worth re-baselining against today.
