Claude Opus 5 Launches With Better Coding and Agentic Performance

Claude Opus 5 is now available as Anthropic’s latest frontier AI model for coding, scientific research, agentic workflows, and complex knowledge tasks.

The new model replaces Claude Opus 4.8 without changing the standard API price. Anthropic says Opus 5 delivers better results while reducing the cost required to complete demanding tasks.

Claude Opus 5 Handles Long-Running Tasks More Effectively

Anthropic designed Claude Opus 5 to work more thoroughly on tasks that require multiple steps or extended periods of execution.

The model can inspect its output, verify whether a solution works, and continue making changes until it reaches a usable result. It also focuses more heavily on identifying the root cause of a problem instead of addressing only visible symptoms.

This approach should make Opus 5 more useful for autonomous coding agents, research systems, and workflows that require repeated testing and correction.

Claude Opus 5 Benchmark Results

Claude Opus 5 scored 43.3% on Frontier-Bench v0.1, more than twice the result achieved by Claude Opus 4.8.

On CursorBench 3.2, the model finished within 0.5 percentage points of Claude Fable 5 while costing approximately half as much.

Opus 5 also surpassed Fable 5 on OSWorld 2.0 while operating at slightly more than one-third of the cost.

Anthropic claims the model outperforms both Fable 5 and GPT-5.6 Sol across several benchmark tests. Results will vary depending on the task, prompt, and tools available to the model.

Performance Approaches Claude Fable 5 at a Lower Cost

Anthropic positions Opus 5 as a more efficient alternative to Claude Fable 5 for many complex workloads.

The company says the model approaches Fable 5 performance while costing around 50% less. It also completes tasks more efficiently than Opus 4.8, lowering the effective cost per successful result.

These improvements could benefit developers running coding agents or other automated systems where models may generate and process large numbers of tokens before completing a task.

Claude Opus 5 Pricing

Claude Opus 5 retains the same standard API pricing as its predecessor:

• Input tokens cost $5 per million tokens.
• Output tokens cost $25 per million tokens.
• Fast mode runs approximately 2.5 times faster.
• Fast mode costs twice the standard base price.

The optional Fast mode targets workloads where response speed matters more than reducing inference costs.

Claude Opus 5 Availability

Claude Opus 5 is available through Anthropic’s applications and API.

The model becomes the default option for Claude Max subscribers. Claude Pro users can also access it as the most capable model included with their subscription.

Developers can select Opus 5 through the API for coding systems, research tools, AI agents, and other complex applications.

Known Claude Opus 5 Limitations

Claude Opus 5 does not lead every evaluation category.

The model still trails Mythos 5 in offensive cybersecurity tasks and some biological research benchmarks. These results suggest that specialized models may retain an advantage in certain high-complexity scientific and security workloads.

However, Opus 5 offers a broader balance of performance, cost, and task-completion reliability compared with Claude Opus 4.8.