Claude Sonnet 5 pricing: the September increase was cancelled
Originally published 19 July 2026 · Updated 28 September 2026
Anthropic's Sonnet 5 launch announcement was updated on August 10 to make the introductory $2/$10 price permanent. The current Claude API pricing documentation lists the same rates. Any forecast that assumed a September increase should be corrected rather than carried forward as a cost event.
Current Sonnet rates
| Model | Input per 1M tokens | Output per 1M tokens | Cache read |
|---|---|---|---|
| Claude Sonnet 5 | $2.00 | $10.00 | $0.20 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $0.20 |
Anthropic's September 28 Sonnet 5.5 announcement lists the same base rates as Sonnet 5, plus $2.50 per million for 5-minute cache writes and $4.00 for 1-hour cache writes. Check the pricing page before procurement or production forecasts, since provider pricing can change.
What Sonnet 5.5 changes for the cost model
Sonnet 5.5 does not lower the published per-token price. Anthropic says the model generates output more than 30% faster and can cost up to 30% less per task for most work because it uses fewer tokens. Treat that as a launch claim and test it against your own workload. Faster output can reduce latency without reducing a token-metered bill; lower spend requires lower token use, fewer retries, or fewer escalations.
For each workload, track cost per accepted result, not just cost per request. Include input/output tokens, cache writes and reads, retries, tool calls, and any fallback to another model. Keep quality and human review rates in the comparison so a cheaper but less reliable run does not appear to be a saving.
Recommended forecast update
- Remove the planned 50% Sonnet 5 rate increase from September and Q4 scenarios.
- Keep Sonnet 5.5 at the same per-token rate in the base case as Sonnet 5.
- Use a separate sensitivity case for reduced tokens per task; do not assume Anthropic's maximum 30% saving applies uniformly.
- Benchmark representative tasks and update the sensitivity with measured cost per accepted result.
For launch details and an example task-cost model, read Sonnet 5.5 pricing: the $2/$10 rate and real task cost and the Sonnet 5.5 model-routing guide.
FAQ
Did Claude Sonnet 5's price increase happen on September 1, 2026?
No. Anthropic cancelled the planned increase. Sonnet 5 remains $2 per million input tokens and $10 per million output tokens.
Does Sonnet 5.5 have a lower API rate than Sonnet 5?
No. Sonnet 5.5 launched at the same listed $2 input and $10 output per million tokens. Anthropic says it can use fewer tokens per task, lowering cost per task by up to 30% for most work.
What should I change in my budget?
Remove the cancelled September price increase from forecasts. Model Sonnet 5.5 savings only after representative tests establish cost per accepted task for your workloads.