Anthropic released Claude Opus 5 on July 24, 2026, making the new model available the same day across all of the company’s platforms. Anthropic describes Opus 5 as a thoughtful and proactive model that approaches the frontier intelligence of Claude Fable 5 at roughly half the price. The launch lands alongside benchmark results from coding, knowledge work, and science evaluations, plus a set of alignment and safety findings reported by Anthropic.
What changed compared with Opus 4.8
Opus 5 keeps the same list pricing as Opus 4.8: 5 dollars per million input tokens and 25 dollars per million output tokens. Anthropic says the model is materially better at verifying its own work and iterating on it, ships with enhanced visual output, and shows what Anthropic and early-access partners describe as stronger judgment and more consistent reasoning.
A fast mode is also available. Anthropic states that fast mode runs about 2.5 times faster than the base configuration at roughly double the cost.
Benchmark results
Anthropic reports the following results for Opus 5 across coding, knowledge work, and science tasks. All figures are taken from Anthropic’s announcement.
| Benchmark | Claude Opus 5 result | Comparison |
|---|---|---|
| Frontier-Bench v0.1 | State of the art | More than doubles Opus 4.8 performance at lower cost |
| CursorBench 3.2 (max effort) | Within 0.5 percent of Fable 5 | Half the cost per task; best result at a given cost across high, xhigh, and max effort |
| FrontierCode 1.1 | Approaches Fable-level performance | Half the cost |
| AA Coding Agent Index | State of the art | Outperforms competitors |
| ARC-AGI 3 | Three times higher than the next-best model | Knowledge work and problem solving |
| Zapier AutomationBench | 1.5 times the pass rate of the next-best model at the same cost; at lowest effort passes more tasks than any other model | Knowledge work and problem solving |
| OSWorld 2.0 | Outperforms all models at one-third the cost of the Fable 5 result | Knowledge work and problem solving |
| GDPval-AA v2 | Best performance at a given cost | Knowledge work and problem solving |
| HLEAutomationBench | Best and cost-efficient performance | Knowledge work and problem solving |
| DeepSearchQA | Best and cost-efficient performance | Knowledge work and problem solving |
| Organic chemistry tasks | 10.2 percentage points higher than Opus 4.8 | Science |
| Protein sequence analysis | 7.7 percentage points higher than Opus 4.8 | Science |
| Life sciences evaluations | Better than Opus 4.8 on every evaluation | Science |
| Box data analysis workflows | 11 percent improvement over Opus 4.8 | Science / domain specific |
| Box due diligence workflows | 17 percent improvement over Opus 4.8 | Science / domain specific |
| Box overall workflows | 8 percent improvement over Opus 4.8 | Science / domain specific |
| Financial modeling | 9 percentage points more accurate on average than Opus 4.8 | Uses one-third fewer turns and 60 percent less time |
| Trading benchmark | Strongest Opus model | Uses one-seventh the reasoning tokens of Opus 4.8 and under half the latency |
What does the coding benchmark performance look like?
On coding evaluations, Opus 5 sets what Anthropic calls state-of-the-art results on Frontier-Bench v0.1 and the AA Coding Agent Index. Anthropic reports the model more than doubles Opus 4.8’s performance on Frontier-Bench v0.1 at lower cost. On CursorBench 3.2 at max effort, Opus 5 lands within 0.5 percent of Fable 5 while costing half as much per task, and is described as the best result at a given cost across high, xhigh, and max effort levels. On FrontierCode 1.1, the model approaches Fable-level performance at half the cost.
How does Opus 5 perform on knowledge work and problem solving?
On knowledge-work and agentic evaluations, Anthropic says Opus 5 scores three times higher than the next-best model on ARC-AGI 3. On Zapier AutomationBench, the model achieves 1.5 times the pass rate of the next-best model at the same cost, and at its lowest effort setting it passes more tasks than any other model tested. OSWorld 2.0 results show Opus 5 outperforming all models at one-third the cost of Fable 5’s result. The model also reports the best performance at a given cost on GDPval-AA v2, and the best and most cost-efficient performance on both HLEAutomationBench and DeepSearchQA.
What did Anthropic report for science and domain-specific tasks?
Anthropic reports several science and domain gains over Opus 4.8. Organic chemistry tasks improved by 10.2 percentage points and protein sequence analysis by 7.7 percentage points. Life sciences evaluations improved on every test. On Box workflows, data analysis improved 11 percent, due diligence improved 17 percent, and overall workflows improved 8 percent. Financial modeling is 9 percentage points more accurate on average than Opus 4.8 while using one-third fewer turns and 60 percent less time. On a trading benchmark, Opus 5 is the strongest Opus model Anthropic has tested and uses one-seventh the reasoning tokens of Opus 4.8 with under half the latency.
How is Opus 5 priced and where is it available?
List pricing matches Opus 4.8: 5 dollars per million input tokens and 25 dollars per million output tokens. A faster mode costs roughly twice as much as the base configuration while running about 2.5 times faster. Anthropic made the model available across all of its platforms on the day of release, July 24, 2026.
What did Anthropic report on alignment and safety?
Anthropic reports an automated behavioral audit score of 2.3 for misaligned behavior, the lowest among recent models in its evaluation. Anthropic states Opus 5 has the lowest rates of deceptive behavior compared with Opus 4.8, Sonnet 5, and Fable 5, and is the safest model Anthropic tested in the reckless-actions risk category.
On the OSS-Fuzz cybersecurity evaluation, Opus 5 is similar to Mythos 5 at vulnerability identification but far behind Mythos 5 at exploit development. Anthropic also notes that Opus 5 remains behind Mythos 5 in biology research and offensive cybersecurity, and does not advance the frontier in dual-use risky capabilities.
Anthropic reports that cyber classifiers intervene about 85 percent less often than they do for Fable 5. A Cyber Verification Program is available for enterprises and researchers.
FAQ
When did Anthropic release Claude Opus 5?
Anthropic released Claude Opus 5 on July 24, 2026, with availability across all Anthropic platforms on the same day.
How much does Claude Opus 5 cost?
Pricing is 5 dollars per million input tokens and 25 dollars per million output tokens, the same as Opus 4.8. A fast mode runs about 2.5 times faster at roughly double the base cost.
What are the main benchmark results for Claude Opus 5?
Anthropic reports state-of-the-art results on Frontier-Bench v0.1 and the AA Coding Agent Index, within 0.5 percent of Fable 5 on CursorBench 3.2 at half the cost, three times higher than the next-best model on ARC-AGI 3, and 1.5 times the pass rate of the next-best model at the same cost on Zapier AutomationBench. Science results include 10.2 percentage points higher than Opus 4.8 on organic chemistry and 7.7 percentage points higher on protein sequence analysis.
Related coverage
Related coverage
This article summarizes reporting from anthropic.com. See our editorial disclaimer for how our articles are produced.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.