Anthropic releases Claude Opus 5, a proactive model close to Fable 5 at half the price
Anthropic released Claude Opus 5 on July 24, 2026, making the new model available the same day across all of its platforms. The company describes Opus 5 as a thoughtful and proactive system that comes close to the frontier intelligence of Claude Fable 5 at roughly half the price. Anthropic is positioning it as the new default on Claude Max and the strongest model on Claude Pro, and as a daily-driver replacement that improves on Opus 4.8 at the same cost.
How does Opus 5 perform on coding and knowledge-work benchmarks?
On coding and knowledge-work evaluations, Anthropic reports state-of-the-art results from Opus 5, with one notable exception. On cybersecurity tasks, Opus 5 remains behind Mythos 5. The strongest numbers come from Frontier-Bench v0.1, where Opus 5 surpasses every other model and more than doubles Opus 4.8’s score at a lower cost per task. On CursorBench 3.2 at maximum effort, the model performs within 0.5 percent of Fable 5’s peak score while charging half as much per task, and at the high, xhigh, and max effort settings it posts a higher score than every other model at the same cost.
Knowledge-work benchmarks tell a similar story. On ARC-AGI 3, an evaluation that asks models to solve novel problems, Opus 5’s score is roughly three times the next-best model. On Zapier AutomationBench, which measures end-to-end completion of business tasks, the pass rate is about 1.5 times the next-best model for the same cost per task, and even at its lowest effort setting Opus 5 passes more tasks than any other model tested. On OSWorld 2.0, a computer-use benchmark, Opus 5 outperforms every other model at any given cost and surpasses Fable 5’s best result at just over a third of the cost.
Opus 5 also leads on several related evaluations that Anthropic highlights, including GDPval-AA v2, HLE, AutomationBench, and DeepSearchQA.
What gains does Opus 5 offer for scientific research?
Anthropic calls Opus 5 a meaningful improvement over Opus 4.8 for scientific research. On its life-sciences evaluations, which span structural biology, organic chemistry, and bioinformatics, the model beats Opus 4.8 on every test. The largest jump is on organic chemistry: Opus 5 scores 10.2 percentage points higher than Opus 4.8 on an internal benchmark that asks the model to infer molecular structures from spectroscopy data. On protein-related tasks, including predicting how sequence variations affect function, it improves by 7.7 percentage points.
How well does Opus 5 generate visual outputs?
According to Anthropic, Opus 5 produces notably stronger visual outputs than its predecessors. The launch post includes two examples: a wind tunnel visualization that shows airflow around aerodynamic and non-aerodynamic objects, and an interactive, simplified illustration of a cell whose individual elements viewers can explore.
How does Opus 5 verify its own work?
Anthropic emphasizes that Opus 5 is much stronger at verifying its work and iterating carefully until it succeeds. Three concrete examples were shared:
- On a Frontier-Bench task in which Opus 5 was given a drawing of a machine part and asked to rebuild it as a 3D FreeCAD model, the model had no direct way to view the image. It responded by writing its own computer vision pipeline to extract geometry from raw pixels, then reconstructed the full machine part, succeeding repeatedly where competing models failed within five attempts.
- Given a real bug in a popular open-source package manager, Opus 5 found the root cause and fixed an edge case that the community’s patch had missed. A competing model fixed only the surface symptom and reported the bug resolved.
- An engineer at a trading firm used Opus 5 to build a market data feed for a new exchange in a single session. Previous models could not complete the task even with extensive plans. With no live feed to validate against, Opus 5 built its own test harness to verify that its code parsed the exchange’s data correctly.
What are customers saying about Opus 5?
Early-access customers highlighted different strengths. Scott Wu, CEO of Devin, said Opus 5 on FrontierCode 1.1 approaches Fable-level performance at half the cost and shows particular strength on difficult debugging and root-cause analysis. Sualeh Asif, co-founder of Cursor, described near-Fable 5 intelligence at Opus speed and cost, with similar behaviors on CursorBench. Wade Foster, CEO of Zapier, said Opus 5 topped Zapier’s AutomationBench leaderboard without spending more tokens than prior Claude models, taking a raw account-health workbook through a full churn-prevention sequence end to end and hitting 100 percent pass rate where previous models did not pass.
Other customer comments covered genomics analysis, financial research, legal work, frontend testing, and IDE integration. Alfredo Andere, CEO, said the model behaves more like a careful scientist, choosing the right statistical tests and cross-checking its own results on genomics work. Ben Kus, CTO of Box, reported that Opus 5 outperforms Opus 4.8 by 8 percent, with 11 percent improvement on data analysis and 17 percent on due diligence workflows. Richard Pham, Evals and Product Lead, said Opus 5 averaged 9 percentage points higher accuracy with a third fewer turns and tool calls and 60 percent less time on financial-modeling tasks. Matt Nassr, Head of Global Data Engineering and AI Transformation, called Opus 5 the strongest Opus model on a trading benchmark, reached with roughly a seventh of the reasoning tokens and under half the latency of Opus 4.8.
Several customers pointed to better judgment and self-checking. AJ Orbach said the model opens its pages in a browser at desktop and phone widths, catches mobile-fold and off-screen button issues, and fixes them before handing work back. Igor Ostrovsky said the model verifies branches, checks templates, and thinks through test implications during pull request handoffs. Denis Shiryaev, Head of AI in IDE at JetBrains, said the model’s judgment is what stands out: it thinks harder before writing, catches logical faults during planning, and reasons about why an answer is right.
How is Opus 5 priced?
Opus 5 is priced at 5 dollars per million input tokens and 25 dollars per million output tokens, the same pricing as Opus 4.8. A fast mode runs about 2.5 times faster at double the base price.
What is the alignment and safety profile of Opus 5?
Anthropic’s pre-deployment automated behavioral audit found Opus 5 to be the company’s most aligned model to date. It adheres to Claude’s Constitution better than Opus 4.8, Sonnet 5, or Fable 5, exhibits the lowest rates of deceptive behavior among recent models, and is the least susceptible to being tricked into misuse. It also scored lowest on reckless actions that could have hard-to-reverse side effects, at 2.3 on overall misaligned behavior.
On dual-use capabilities, Anthropic says Opus 5 does not advance the frontier in risky areas. In evaluations run with private-sector and government partners, it remains behind Mythos 5 in both biology research and offensive cybersecurity. As with Opus 4.8, Anthropic intentionally avoided training Opus 5 on cyber tasks, yet the model improved on those tasks as it became more generally capable. On OSS-Fuzz, an internal evaluation that measures both finding and exploiting vulnerabilities without extensive human guidance, Opus 5 and Mythos 5 identify vulnerabilities with similar success, but Opus 5’s score on developing exploits is far behind Mythos 5’s.
FAQ
What is Claude Opus 5 and when was it released?
Claude Opus 5 is Anthropic’s new flagship model, announced and made available on July 24, 2026 across all Anthropic platforms. Anthropic describes it as a thoughtful, proactive model that approaches the frontier intelligence of Claude Fable 5 at roughly half the price.
How much does Claude Opus 5 cost?
Opus 5 is priced at 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 double the base price.
How does Opus 5 compare to Fable 5 and Mythos 5?
Anthropic reports that Opus 5 comes close to Fable 5 on intelligence benchmarks while costing about half as much, and that on several evaluations it approaches or surpasses Fable 5’s results at a fraction of the cost. On cybersecurity tasks, Opus 5 remains behind Mythos 5 in both identifying and exploiting vulnerabilities, although it identifies them at a similar rate.
Related coverage
- Anthropic Settles Claude Fable 5 Access, Keeps Model Inside Max and Team Premium at Half Capacity
- White House accuses Moonshot AI of distilling Anthropic’s Fable for Kimi K3
This article summarizes reporting from anthropic.com.