Anthropic announced the release of Claude Opus 5, a model the company describes as approaching the frontier intelligence of Claude Fable 5 at roughly half the price. On coding and knowledge work evaluations including Frontier-Bench and GDPval-AA, Opus 5 sets a new state-of-the-art among Anthropic’s models, though it remains behind Mythos 5 on cybersecurity tasks. Opus 5 is now the default model on Claude Max and the strongest model available on Claude Pro.
Anthropic said Opus 5 delivers substantially improved performance for the same cost as its predecessor, Opus 4.8. On Frontier-Bench v0.1, the company said Opus 5 surpasses all other models and more than doubles Opus 4.8’s performance at a lower cost per task, while on CursorBench 3.2 at maximum effort it performs within 0.5 percent of Fable 5’s peak score at half the cost per task. On ARC-AGI 3, an evaluation involving novel problem-solving, Opus 5’s score is three times that of the next-best model, and on Zapier’s AutomationBench, which measures end-to-end completion of business tasks, its pass rate is roughly 1.5 times higher than the next-best model at the same cost. On OSWorld 2.0, a computer use benchmark, the company said Opus 5 outperforms every other model at any given cost and surpasses Fable 5’s best result at just over a third of the cost. Anthropic also said Opus 5 shows meaningful improvement over Opus 4.8 on life sciences evaluations spanning structural biology, organic chemistry, and bioinformatics, with the largest gains on tasks such as inferring molecular structures from spectroscopy data and predicting how protein sequence variations affect function.
On alignment, Anthropic said its automated behavioral audit found Opus 5 to be its most aligned model to date, scoring 2.3 on overall misaligned behavior, the lowest of its recent models, with the lowest rates of deceptive behavior and the least susceptibility to being tricked into misuse among models tested. On safety, the company said Opus 5 does not advance the frontier in risky, dual-use capabilities, remaining behind Mythos 5 in both biology research and offensive cybersecurity based on evaluations conducted with private-sector and government partners. The model was not intentionally trained on cyber tasks, though Anthropic said it has improved on cybersecurity evaluations as a result of general capability gains, coming close to Mythos 5 at identifying software vulnerabilities on the company’s OSS-Fuzz evaluation while remaining substantially behind Mythos 5 at developing exploits from those vulnerabilities.
Opus 5’s cyber classifiers are proportionally less restrictive than those applied to Fable 5, allowing the model to find vulnerabilities in source code while blocking binary-based vulnerability scanning, penetration testing, and exploit generation; Anthropic said it expects these classifiers to intervene around 85 percent less often than they do for Fable 5, with flagged requests falling back to Opus 4.8 by default in Claude.ai, Claude Code, and Claude Cowork. Members of Anthropic’s Cyber Verification Program have access to a version of Opus 5 with fewer security restrictions. On biology, Opus 5 uses a similar safeguard suite to Opus 4.8, making it Anthropic’s most capable generally available model for scientific research, though the company said it still shows limitations on long-running, autonomous research tasks, an area where Mythos 5 remains the stronger model.
Claude Opus 5 is available today across all platforms, priced at $5 per million input tokens and $25 per million output tokens, the same pricing as Opus 4.8, and is also available in a Fast mode that runs approximately 2.5 times the default speed at twice the base price. Alongside the release, Anthropic introduced two beta features: mid-conversation tool changes on the Claude Platform, which let developers change which tools Claude can use without invalidating the prompt cache, and automatic fallbacks on the API, which allow requests flagged by safety classifiers to route automatically to another available model rather than being blocked.
