Anthropic launches Claude Opus 5 as a more efficient frontier model

Anthropic has launched Claude Opus 5, a frontier model for coding, knowledge work and long-running agent tasks. Announced on July 24, it is available on Anthropic’s platforms and through the Claude API as claude-opus-5. It is the default model for Claude Max and the strongest option included with Claude Pro.

The release focuses on efficiency as well as capability. Anthropic says Opus 5 comes close to its more powerful Fable 5 model on many tasks while costing roughly half as much per completed task. API pricing stays at the Opus 4.8 level: $5 per million input tokens and $25 per million output tokens. A Fast mode runs about 2.5 times faster at twice the base token price.

Users can adjust an effort setting to trade deeper reasoning for speed and lower token use. In Anthropic’s tests, Opus 5 more than doubled Opus 4.8’s result on Frontier-Bench at a lower cost per task. At maximum effort it came within 0.5 percentage points of Fable 5 on CursorBench. Anthropic also reports gains in computer use, business automation, scientific research and visual output. These are provider-run and partner evaluations, so organisations should still test the model on their own workloads.

Reliability is a central claim. Anthropic says Opus 5 verifies its work more carefully, iterates until a task succeeds and is less likely to report completion after fixing only a surface symptom. That matters for coding agents and office workflows, where a confident but incomplete result creates more work. GitHub has already started a gradual rollout in Copilot for Pro+, Max, Business and Enterprise users.

Anthropic says Opus 5 does not push the frontier in the highest-risk biological or offensive cyber capabilities. It approaches Mythos 5 at finding software vulnerabilities but remains substantially weaker at turning them into working exploits. Its safeguards allow source-code vulnerability research while restricting binary scanning, penetration testing and exploit generation. Flagged requests can fall back to Opus 4.8, while verified security researchers can request broader access.

For users, Opus 5 offers a stronger everyday Claude model without a higher list price. For makers and companies, frontier-like performance is moving into a cheaper, adjustable tier that can be used more routinely. The practical decision is no longer simply which model scores highest. Teams must compare cost per finished task, latency, verification quality, safety restrictions and how reliably an agent recovers when its first approach fails.