Claude Opus 5

Claude Opus 5 is a proprietary large language model released by Anthropic on July 24, 2026, the first Opus-tier model of the Claude 5 generation of the Claude family.¹ Anthropic described it as coming "close to the frontier intelligence" of Claude Fable 5 at half the price, with API pricing unchanged from Claude Opus 4.8 at $5 per million input tokens and $25 per million output tokens.¹ At launch it became the default model on the Claude Max plan and the strongest model available on Claude Pro; it was Anthropic's fourth model release in under two months, following Mythos 5, Fable 5, and Claude Sonnet 5.¹⁵

Features

A customer-facing effort setting trades intelligence against token cost and speed: press coverage described a three-level toggle in the consumer products, while the developer platform exposes further high, xhigh, and max settings, defaulting to high on the API and Claude Code.¹²⁵ A fast mode runs at roughly 2.5 times the default speed for twice the base price, as with Opus 4.8.¹ The model carries a 1,000,000-token context window with a 128,000-token maximum output, a stated knowledge cutoff of May 2026, and the API identifier claude-opus-5.² Two platform features shipped alongside the release: developers can change the model's available tools mid-conversation without invalidating the prompt cache, and API requests flagged by safety classifiers can automatically fall back to another model.¹ GitHub made Opus 5 available in Copilot's paid tiers the same day.⁴

Performance

Anthropic's launch materials reported that Opus 5 more than doubles Opus 4.8's score on Frontier-Bench v0.1 at a lower cost per task, performs within 0.5 percent of Fable 5's peak CursorBench 3.2 score at half the cost, triples the next-best model's score on ARC-AGI 3, passes roughly 1.5 times as many Zapier AutomationBench tasks as the next-best model at the same cost, and surpasses Fable 5's best OSWorld 2.0 computer-use result at just over a third of the cost, all developer-reported figures as of July 2026.¹ The company also reported its largest life-sciences gain on an internal organic chemistry benchmark, 10.2 percentage points over Opus 4.8.¹ On launch day the model led the Artificial Analysis leaderboard ahead of Fable 5, and early commentary described the reception as positive.³ A same-day hands-on review praised the model's capability while criticizing its verbosity and what the reviewer called a "neurotic" personality, including a merge conflict it declined to resolve.⁶

Safety

Anthropic reported Opus 5 as its most aligned model to date on an automated behavioral audit, scoring 2.3 on overall misaligned behavior, the lowest of its recent models, with better adherence to Claude's Constitution than Opus 4.8, Sonnet 5, or Fable 5.¹ The company stated the model does not advance the frontier in risky dual-use capabilities: it remains behind Mythos 5 in biology research and offensive cybersecurity, and its cyber classifiers permit source-code vulnerability finding while blocking binary-based scanning, penetration testing, and exploit generation, intervening an expected 85 percent less often than Fable 5's.¹ Flagged requests in Anthropic's consumer products fall back to Opus 4.8 by default. Anthropic called Opus 5 its most capable generally available model for scientific research, and, unlike Fable 5, it carries no data-retention requirement for general access.¹

Position in the lineup

Opus 5 brings the Opus tier into the Claude 5 generation after the 4.x series closed with Claude Opus 4.8. It sits below Claude Fable 5, which remains Anthropic's most capable model and its recommendation for the most advanced autonomous work, while the developer documentation recommends Opus 5 as the default starting model for complex agentic coding and enterprise work.²⁵ Like other recent Claude releases it is a hybrid reasoning model whose parameter count and training corpus are undisclosed, with post-training in the Constitutional AI and RLHF tradition and effort controls governing test-time compute.

References