Claude Sonnet 5.5 runs 30% faster and costs less per task

Claude Sonnet 5.5 is Anthropic's new model, announced on September 28, 2026: it runs over 30% faster than Sonnet 5 and costs up to 30% less per task. That's a speed and cost pitch, with numbers taken from Anthropic's own announcement.

What changes with Claude Sonnet 5.5

According to Anthropic, the model is faster and uses less per task than Sonnet 5. One caveat: the "up to 30% less per task" is measured in tokens per task versus Sonnet 5. It's not an average, and it doesn't mean everything costs 30% less.

API pricing and key numbers

These are the three figures the announcement highlights:

• API price: $2 per million input tokens and $10 per million output tokens.
• Computer use: 80.1% on the OSWorld 2.1 benchmark, a partial score.
• Office work: 1844 points on GDPval-AA, versus 1846 for Opus 5.5. That's a near tie with Anthropic's larger model.

What this model doesn't prove

Every figure comes from Anthropic; none is an independent measurement. The near tie with Opus 5.5 comes from a single evaluation (GDPval-AA), so it applies to office work, not everything. There's also no verified Opus 5.5 price to claim Sonnet 5.5 is cheaper than it, and the official page doesn't say which Claude app plans include it.

How to apply it in your business?

The practical takeaway: instead of using the big model out of habit, test the newer one first on your real task. Pick two or three repetitive tasks on your team, run Sonnet 5.5, and compare result, time and cost against what you use today. If the result is equivalent, you pay less without losing quality; if not, you have the data to justify the pricier model.

For another recent Anthropic move, read the full breakdown of Claude Marketplace.

Frequently asked questions

What is Claude Sonnet 5.5?

It's Anthropic's new model, announced on September 28, 2026, that runs over 30% faster than Sonnet 5.

How much does Claude Sonnet 5.5 cost on the API?

$2 per million input tokens and $10 per million output tokens.

Is Sonnet 5.5 better than Opus 5.5?

Not by these numbers: it nearly ties on office work (1844 vs 1846 on GDPval-AA), but that's a single Anthropic evaluation.

What does "up to 30% less per task" mean?

It's a tokens-per-task measurement versus Sonnet 5. It's a ceiling ("up to"), not an average.

Conclusion

Claude Sonnet 5.5 promises more speed and lower cost per task, with a near tie against Opus 5.5 on office work. Those are Anthropic's own figures, so the sensible move is to test it on your real task before deciding which model to pay for.