Claude Opus 5

What Is Claude Opus 5?

Claude Opus 5 is Anthropic's latest Opus-tier hybrid reasoning model, released July 24, 2026. It sits between Sonnet 5 and Fable 5 in Anthropic's lineup and is positioned as the model you reach for by default when you need frontier-class intelligence without frontier-class costs. It is now the default model on Claude Max and the strongest model available on Claude Pro.

What It's Good For

Agentic coding and software engineering is where Opus 5 is most clearly differentiated. On Frontier-Bench v0.1 — a 74-task agentic terminal coding benchmark — Opus 5 scored 43.3% at max effort, more than doubling Opus 4.8's score of 18.7% and outperforming both Fable 5 (33.7%) and GPT-5.6 Sol (37.5%). On CursorBench 3.2, it came within 0.5% of Fable 5's peak score at half the per-task cost, making it a practical choice for teams running GitHub Copilot-style developer workflows at scale.

Computer use and automation are another strong area. On OSWorld 2.0, Opus 5 surpasses Fable 5's best result at just over one-third of the cost. On Zapier AutomationBench, it achieved roughly 1.5× the pass rate of the next-best model at the same cost per task — a meaningful signal for teams building business-process automation pipelines.

Scientific and knowledge work also benefits from the model's improvements. Opus 5 outperforms Opus 4.8 on every life sciences evaluation Anthropic tracks, with a 10.2-percentage-point gain on organic chemistry tasks involving molecular structures inferred from spectroscopy data, and a 7.7-percentage-point gain on protein-function prediction. Its knowledge cutoff of May 2026 — the freshest in the entire Claude lineup — also makes it more current than comparable models for tasks involving recent tooling, APIs, or rapidly evolving research.

For teams building long-context reasoning pipelines, the 1M-token context window (with 128K max output on the standard API) enables document analysis, large-codebase review, and multi-step agent tasks that would overflow smaller context models. Unlike Fable 5, Opus 5 also has no data retention requirements, which matters for enterprises with strict data-handling policies.

Who It's a Good Fit For

• Engineering teams running agentic scaffolds (Claude Code, custom agent harnesses) who need strong coding performance without paying Fable 5 rates
• Enterprises with zero-data-retention requirements that still need top-tier model capability
• Researchers working in life sciences, chemistry, or data-dense domains who benefit from the May 2026 knowledge cutoff
• Developers already using Claude 3 Opus who want a direct, drop-in upgrade at unchanged token pricing

For teams comparing AI writing or content generation tools like Perplexity for research workflows, Opus 5's stronger factual recall and longer context may justify the API cost for high-stakes outputs.

You can explore the official model page at anthropic.com/claude/opus.

Limitations and Where It Falls Short

Opus 5 is not the undisputed leader across every benchmark. On DeepSWE v1.1, it scores 68.8, behind Fable 5 (69.7) and GPT-5.6 Sol (72.7) — a real gap for teams where that specific benchmark reflects their workload.

The web fetch tool is not available on Opus 5, which limits its usefulness for real-time web retrieval tasks. Additionally, while accuracy is reported as 11% higher than Opus 4.8, the hallucination rate is also 6% higher — meaning the model answers more questions but introduces more false claims alongside correct ones. Pipelines that require high factual precision should account for this trade-off.

Finally, cybersecurity offensive tasks are intentionally restricted. Anthropic has added stronger safeguards around exploit generation and binary-based vulnerability scanning, and requests flagged by safety classifiers automatically fall back to Opus 4.8.