Opus 5

2026-07-24 · Source: AINews · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, extended

Summary

Anthropic launched its Claude Opus 5 model, sparking mixed reactions regarding its performance and evaluation. The model achieved an Epoch Capabilities Index (ECI) of 159, slightly below Fable 5's 161, but matched Fable 5 with a SWE-ECI of 161 on software engineering benchmarks. This result drew criticism, with some users calling Opus 5 "incredibly underrated" given its practical improvements over Opus 4.8, which scored only 1 point lower. An anomaly on FrontierCode showed Opus 5 performing better at medium effort than high effort. Anecdotal evidence from Microsoft CTO Kevin Scott and others praised Opus 5's coding and math capabilities, especially with "best-of-n" sampling, and its agentic tool use, including browser automation. Nous Research also provided access to Opus 5 with a 20% discount.

Key takeaway

For Machine Learning Engineers evaluating frontier models for agentic applications, consider that Claude Opus 5's real-world coding and browser automation capabilities may surpass its initial ECI benchmarks. You should prioritize practical, task-specific evaluations over aggregate scores, especially when using "best-of-n" sampling. Also, adapt your prompt engineering to the new progressive disclosure guidelines for Claude 5 to optimize performance and avoid over-constraining the model.

Key insights

Claude Opus 5's practical coding and agentic capabilities appear to exceed its initial aggregate benchmark scores, highlighting evaluation challenges.

Principles

• Aggregate benchmarks may understate real-world model utility.
• "Best-of-n" sampling can significantly improve model outcomes.
• Increased inference effort doesn't always guarantee performance gains.

Method

Progressive disclosure in prompt engineering minimizes persistent context, loading detailed rules only when relevant for Claude 5 models.

In practice

• Test Opus 5 for browser automation and long-horizon coding tasks.
• Implement "best-of-n" sampling for improved Opus 5 results.
• Audit system prompts for older Claude models using the "/doctor" command.

Topics

• Claude Opus 5
• Large Language Models
• Model Benchmarking
• Agentic AI
• Software Engineering
• Prompt Engineering
• Browser Automation

Code references

• danilo-znamerovszkij/draw-your-font

Best for: AI Engineer, CTO, VP of Engineering/Data, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• [AINews] Claude Opus 5: Fable-level performance at Opus price (half Fable) — Claude Opus 5 demonstrates advanced agentic coding and tool-use, challenging the adequacy of current aggregate benchmarks.
• not much happened today — AI models are advancing in design, coding, and agentic capabilities, with a focus on efficiency, local deployment, and robust evaluation.
• OPUS 4.8!!! (also maybe GPT5.6??) — Opus 4.8 enhances AI agentic capabilities and cost-efficiency through parallel processing and optimized inference.
• Claude Opus 5: Near-Frontier Intelligence, On a Dial — Claude Opus 5 delivers near-frontier intelligence with enhanced reasoning and context at its predecessor's price point.
• Claude Opus 5 Is THE GREATEST AI Model EVER?! Beats Fable & CHEAPER! (Fully Tested) — Claude Opus 5 offers frontier-class AI capabilities at significantly lower cost than competitors, excelling in complex reasoning and autonomous code generation.
• Claude Opus 5 Is #1 Overall. It Still Loses at the One Thing I Run Daily. — Overall benchmark leadership does not guarantee superiority for specific, common daily workflows.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AINews.