Anthropic has made Claude Opus 5 available on all platforms. The company positions it as approaching the frontier intelligence of Claude Fable 5 at half the price, while remaining behind Mythos 5 on cybersecurity tasks. It becomes the default model on Claude Max and the strongest option on Claude Pro.

Pricing stays at $5 per million input tokens and $25 per million output tokens, identical to Opus 4.8. A Fast mode runs at roughly 2.5 times the default speed for twice the base price on the Claude Platform and via usage credits in Claude Code. An effort setting lets users trade intelligence against speed and token cost.

On benchmarks, Anthropic reports that Opus 5 leads Frontier-Bench v0.1 and more than doubles Opus 4.8's performance at a lower cost per task. On CursorBench 3.2 at max effort it comes within 0.5% of Fable 5's peak at half the cost. Its ARC-AGI 3 score is said to be three times the next-best model, its Zapier AutomationBench pass rate about 1.5x the nearest competitor, and on OSWorld 2.0 it beats Fable 5's best result at just over a third of the cost. Life-science evaluations improved across the board, notably +10.2 percentage points on an internal organic chemistry benchmark and +7.7 points on protein variant prediction.

Anthropic highlights agentic behavior from testing: Opus 5 wrote its own computer vision pipeline to reconstruct a 3D FreeCAD machine part from an image it could not directly view; it fixed the root cause of a real open-source package manager bug that a community patch had missed; and it built a market data feed for a new exchange in one session, writing its own test harness for validation. Early-access customers including Cursor, Zapier, Lovable, JetBrains, Box and others report gains, with Lovable citing 22% improvement on hard agentic coding tasks and one financial firm reporting 9 points higher accuracy with a third fewer turns and 60% less time.

On safety, Anthropic's behavioral audit scored Opus 5 at 2.3 for overall misaligned behavior, the lowest of its recent models. It trails Mythos 5 in biology research and offensive cybersecurity, and was deliberately not trained on cyber tasks, though general capability gains brought it close to Mythos 5 at finding vulnerabilities while staying well behind on exploit development. Cyber classifiers are less restrictive than Fable 5's, intervening an expected 85% less often; flagged requests in Claude.ai, Claude Code and Claude Cowork fall back to Opus 4.8 by default. Cyber Verification Program members get immediate access to a less restricted version.

Two beta features ship alongside: mid-conversation tool changes on the Claude Platform without invalidating the prompt cache, and automatic API fallbacks that route safety-flagged requests to another model. Developers can use the model as claude-opus-5 on the Claude API, with no data retention requirements for general access.