The bottom line: Opus 5 delivers partly better performance than Claude 5 on software development tasks and computer-use scenarios, but costs significantly less and is priced the same as Opus 4.8.
Anthropic has released Claude Opus 5 and positions the model as a cost-effective alternative to the expensive flagship model Claude 5. In several software development and problem-solving benchmarks, Opus 5 outperforms the competition, in some cases even matching the performance of Claude 5 — at half the cost.
Claude Opus 5 is now available on all platforms and costs 5 US dollars per million input tokens and 25 US dollars per million output tokens — identical to its predecessor Opus 4.8. A fast mode with 2.5x speed costs double the base price. API integration uses the designation claude-opus-5.
On software development benchmarks, Opus 5 shows significant progress. On Frontier Bench v0.1, the model doubles the score of Opus 4.8 and outperforms all competitors. In CursorBench 3.2 in maximum effort mode, Opus 5 nearly matches Claude 5’s level at half the cost per task. On computer-use tasks (OSWorld 2.0), Opus 5 even exceeds Claude 5’s top result at roughly one-third of the cost. In the ARC-AGI-3 test, the score is three times higher than the next best model. In Zapier AutomationBench, which measures autonomous completion of business tasks, Opus 5 achieves about 1.5x the success rate of competitors at the same price.
Additionally, Anthropic has demonstrated progress in life sciences. The model improves the derivation of molecular structures from spectroscopy data and predictions of protein functions. This is also demonstrated by a visual simulation of airflow around aerodynamic objects in a virtual wind tunnel, showcasing improved visual capabilities.
According to Anthropic, Opus 5 is the company’s best-aligned model. Automated behavior audits show the lowest values for deceptive behavior and misuse susceptibility of all previous models. Security classifiers intervene around 85 percent less frequently than with Claude 5. Opus 5 can identify vulnerabilities in software source code, but binary-based scanning, penetration testing, and exploit development are blocked — a deliberate choice against targeted training for cyber-attack techniques. Blocked requests fall back to Opus 4.8 by default in Claude.ai, Claude Code, and Claude Cowork.
Anthropic announces two beta features: the ability to switch available tools mid-conversation without losing prompt cache, and automatic API fallbacks that redirect requests blocked by security classifiers to another model automatically.
Source: www.it-daily.net · Published July 27, 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.