Anthropic presents Claude Opus 5.5 as a capable new model for software development, tool use and complex professional tasks. Its most significant advance may be economic: greater efficiency with markedly lower operating costs.

A model designed for long assignments

Anthropic unveiled Claude Opus 5.5 on 22 September 2026. Rather than answering a single question, it targets long assignments such as analysing a large software project, using several tools and working through multiple steps.

Anthropic reports progress in agentic coding, autonomous computer use and complex knowledge work. Benchmark figures must still be read cautiously because settings and test conditions influence them.

Faster and less expensive

Anthropic says Opus 5.5 generates output text more than 30% faster than Opus 5 and cuts the average cost of some work by about 40%.

For businesses, lower computing cost could broaden its use in code audits, application migration, document analysis and internal automation.

More enduring agents

Anthropic reports that an early-access tester audited and fixed a 200,000-line codebase in under three hours. This company-reported result does not replace independent evaluation.

The deeper change is endurance: an agent can inspect a codebase, edit several files, run tests and resume after a failure.

Human oversight remains essential

Greater autonomy also raises risk. An agent with terminal access can propagate an error or act on a mistaken interpretation at scale.

Actions should remain limited, logged and reviewable, with human approval before sensitive operations.

Conclusion

Opus 5.5 illustrates a new stage in the AI race: not just the best score, but a fast, affordable and dependable agent that can work for hours.

If independent tests confirm the claims, operational efficiency may matter more than raw power alone.