Anthropic releases Claude Opus 5, available now across all platforms
Anthropic announced Claude Opus 5 on July 24, 2026, and made the model available the same day across all of its platforms. The company positions Opus 5 as a thoughtful, proactive model that approaches the frontier intelligence of Claude Fable 5 at roughly half the price, while delivering what Anthropic calls a clear improvement over its predecessor, Opus 4.8, at the same cost.
What benchmarks does Opus 5 lead on?
Anthropic’s announcement highlights a series of benchmark results designed to show that Opus 5 advances the state of the art on coding and knowledge-work evaluations, even as it trails another model on cybersecurity tasks.
- Frontier-Bench v0.1: Opus 5 surpasses all other models and more than doubles Opus 4.8’s performance at a lower cost per task.
- CursorBench 3.2: At maximum effort, Opus 5 performs within 0.5% of Claude Fable 5’s peak score at half the cost per task, and outperforms every other model on high, xhigh, and max effort at any given cost.
- ARC-AGI 3: A novel-problem-solving evaluation on which Opus 5 scores roughly three times higher than the next-best model.
- Zapier AutomationBench: Pass rate is about 1.5 times the next-best model for the same cost, and even at its lowest effort setting Opus 5 passes more tasks than any other model.
- OSWorld 2.0: A computer-use benchmark where Opus 5 surpasses Fable 5’s best result at just over a third of the cost.
Opus 5 is also described as Anthropic’s best and most cost-efficient model on ARC-AGI 3, GDPval-AA v2, OSWorld 2.0, HLE, AutomationBench, and DeepSearchQA.
How does Opus 5 improve scientific research and visual output?
Anthropic says Opus 5 is a meaningful improvement over Opus 4.8 for scientific research, with gains across life sciences evaluations covering structural biology, organic chemistry, and bioinformatics. The largest jumps appear in two areas:
- Organic chemistry tasks such as inferring molecular structures from spectroscopy data: Opus 5 scores 10.2 percentage points higher than Opus 4.8 on Anthropic’s internal benchmark.
- Protein-related tasks such as predicting how sequence variations affect function: a 7.7 percentage-point improvement over Opus 4.8.
The company also points to stronger visual output, including an interactive wind tunnel illustration showing airflow over aerodynamic and non-aerodynamic objects, and a simplified, interactive illustration of a cell.
How capable is Opus 5 at verifying and iterating on its own work?
Anthropic emphasizes that Opus 5 is much stronger at verifying its own work and iterating carefully. Three examples are highlighted in the announcement:
- On a Frontier-Bench task that asked Opus 5 to rebuild a machine part as a 3D FreeCAD model from a drawing the model could not directly view, Opus 5 wrote its own computer vision pipeline to extract geometry from raw pixels and reconstructed the part. No competing model with the same setup solved the task in five attempts.
- Given a real bug in a popular open-source package manager, Opus 5 found the root cause and fixed an edge case the community’s patch had missed, while a competing model fixed only the surface symptom.
- An engineer at a trading firm used Opus 5 to build a market data feed for a new exchange in a single session. With no live feed to validate against, Opus 5 built its own test harness to check that its code parsed the exchange’s data correctly.
What do early-access customers say about Opus 5?
Customer testimonials in the announcement span coding, analytics, finance, and legal work:
- Scott Wu, CEO of Devin: On FrontierCode 1.1, Opus 5 approaches Fable-level performance at half the cost and shows particular strength on difficult debugging and root-cause analysis tasks.
- Sualeh Asif, Co-Founder at Cursor: On CursorBench, Opus 5 is just under Fable 5 and shows many of the same behaviors.
- Wade Foster, CEO of Zapier: Opus 5 topped Zapier’s AutomationBench leaderboard without spending more tokens than prior Claude models, and ran a full churn-prevention sequence end to end with a 100% pass rate on a test where previous models did not pass.
- Alfredo Andere, CEO of an unnamed genomics analysis firm: Opus 5 behaves more like a careful scientist, choosing appropriate statistical tests, cross-checking results by independent methods, and staying on track through long multi-step analyses.
- Fabian Hedin, Co-Founder at Lovable: On internal evals Opus 5 came out ahead of every model in its family, including a 22% gain over Opus 4.7 on the hardest agentic coding tasks and far less run-to-run variance.
- Madhav Jha, Co-Founder and CTO at Lovable: Calls Opus 5 the biggest leap in the Opus family since 4.5, with the strongest animations, games, and 3D work seen from an Opus model.
- Ben Kus, CTO at Box: Box found that Opus 5 outperforms Opus 4.8 by 8% overall, with an 11% improvement in data analysis workflows and a 17% improvement in due diligence workflows.
- Richard Pham, Evals and Product Lead at an unnamed firm: On hard financial-modeling tasks, Opus 5 averaged 9 percentage points higher accuracy than Opus 4.8 across effort levels, with a third fewer turns and tool calls and 60% less time.
- Niko Grupen, Head of Applied Research at an unnamed legal-focused firm: Saw the biggest gains in practice areas like corporate governance and arbitration, and noted similar performance at 26% fewer tokens on average compared to Opus 4.8 at max reasoning.
- Igor Ostrovsky, Co-Founder and CTO at an unnamed firm: Plans to migrate use cases in Cosmos, the company’s unified agent platform, including code review.
- Denis Shiryaev, Head of AI in IDE at JetBrains: Highlights Opus 5’s judgment during planning and reasoning about why an answer is right, not just whether it works.
- Matt Nassr, Head of Global Data Engineering and AI Transformation at an unnamed trading firm: Describes Opus 5 as the strongest Opus model on the firm’s trading benchmark, reaching that level with roughly a seventh of the reasoning tokens and under half the latency of Opus 4.8.
- Deepak Singh, VP of Agentic AI at Kiro: Calls Opus 5 a strong agentic coding model for long-running, multi-step work that holds the thread across complex tasks.
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. According to the announcement, it adheres to Claude’s Constitution better than Opus 4.8, Sonnet 5, or Fable 5; exhibits the lowest rates of deceptive behavior; and is the least susceptible to being tricked into misuse. On overall misaligned behavior, the audit gave Opus 5 a score of 2.3, the lowest of Anthropic’s recent models.
On safety, the company states that Opus 5 does not advance the frontier in risky, dual-use capabilities. In evaluations conducted with private-sector and government partners, Opus 5 remains behind Mythos 5 on both biology research and offensive cybersecurity. Anthropic notes that, as with Opus 4.8, it intentionally avoided training Opus 5 on cyber tasks; the model nonetheless improved on these tasks as a general capability gain and comes close to Mythos 5 at finding vulnerabilities, while remaining substantially behind Mythos 5 on exploiting them. The announcement references OSS-Fuzz as an evaluation that illustrates this gap.
How is Opus 5 priced, and where is it available?
Pricing for Opus 5 is set at $5 per million input tokens and $25 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. Opus 5 is available today across all platforms and is the new default model on Claude Max and the strongest model on Claude Pro.
FAQ
When was Claude Opus 5 released?
Anthropic announced Claude Opus 5 on July 24, 2026, and made the model available the same day across all platforms.
How does Opus 5 compare to Claude Fable 5?
According to Anthropic, Opus 5 approaches the frontier intelligence of Claude Fable 5 at roughly half the cost. On CursorBench 3.2 at max effort, Opus 5 performs within 0.5% of Fable 5’s peak score at half the cost per task, and on OSWorld 2.0 it surpasses Fable 5’s best result at just over a third of the cost.
What does Opus 5 cost per million tokens?
Opus 5 is priced at $5 per million input tokens and $25 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.
Related coverage
- Anthropic Ends Claude Fable 5 Subscription Limbo: Permanent Max and Team Premium Caps, With Lower Plans Moving to Pay-Per-Use
- Anthropic describes an internal reasoning space inside Claude and stops short of calling it conscious
This article summarizes reporting from anthropic.com.