If you’re running production workloads on Claude Sonnet 5, your bill is going up on September 1, 2026, whether or not your usage changes at all. Anthropic’s introductory pricing for the model expires August 31, and the standard rate that follows is a flat 50% increase on both input and output tokens.
Quick facts
- Claude Sonnet 5 launched June 30, 2026 at introductory pricing of $2 per million input tokens and $10 per million output tokens.
- That introductory pricing expires August 31, 2026; standard pricing of $3 per million input tokens and $15 per million output tokens begins September 1.
- The standard rate matches Claude Sonnet 4.6’s list price exactly, but Sonnet 5 uses a new tokenizer that counts roughly 1.0 to 1.35 times more tokens for the same text.
- Sonnet 5 is the default model for Free and Pro users on claude.ai, and is available in Claude Code, the Claude API, Cursor, VS Code, and GitHub Copilot.
Why “same list price” doesn’t mean “same bill”
Per Anthropic’s own launch announcement, Sonnet 5’s standard per-token rate is identical to what Sonnet 4.6 already charges. On paper, that reads like no change at all for anyone already budgeting at the 4.6 rate. The catch is the tokenizer: because Sonnet 5’s tokenizer breaks the same text into more tokens than 4.6’s did, particularly for code, structured data, and non-English text, a workload that looks stable on the rate card can still cost meaningfully more per request once the introductory discount disappears. Independent cost analysis has put the real-world token increase as high as 35% for token-heavy content types.
What actually changes on your bill September 1
The rate change alone is a 50% jump: $2 to $3 on input, $10 to $15 on output. Layer the tokenizer effect on top and a workload processing 500 million input tokens and 150 million output tokens a month could see a monthly increase in the range of $1,000 to $1,500, compounding to well over $10,000 a year, depending on how token-dense your specific content is. That applies whether you’re calling the API directly, or indirectly through AWS Bedrock or Google Vertex AI, both of which pass Anthropic’s pricing through on the same September 1 timeline.
What to actually do before August 31
Three concrete moves are worth making in the next few weeks. First, re-benchmark your actual token usage under the new tokenizer rather than assuming your old 4.6-era estimates still hold, since that’s the number the rate hike actually multiplies against. Second, look at prompt caching, which cuts cache-hit input costs by 90% and can absorb a meaningful share of the increase for workloads that reuse the same context repeatedly. Third, for tasks that don’t need Sonnet-tier reasoning, evaluate shifting to Claude Haiku 4.5 specifically for the highest-volume, lowest-complexity parts of your pipeline, since that’s where the cost difference compounds fastest at scale.
Key takeaway
August 31 is a hard deadline, not a soft target, per Anthropic’s own published pricing page, and no extension has been announced. If Claude Sonnet 5 is anywhere in your production stack, the next few weeks are the window to re-benchmark actual token counts, check whether prompt caching applies to your workload, and budget for a 50%-plus increase rather than being surprised by the September invoice.
Leave a Reply