Claude Sonnet 5's launch pricing ends August 31 — the real bill increase is bigger than the sticker price
Anthropic's launch pricing for Claude Sonnet 5 — $2/$10 per million input/output tokens — ends 31 August 2026 and reverts to standard $3/$15 pricing on 1 September, a flat 50% increase; but Sonnet 5's new tokenizer also generates roughly 30% more tokens for the same work, so some teams' effective bill will rise by well over 50%.
6 August 2026
Anthropic’s introductory pricing for Claude Sonnet 5 — $2 per million input tokens, $10 per million output tokens — expires on 31 August 2026. From 1 September, standard pricing of $3/$15 applies, a straightforward 50% increase on both input and output. That much was flagged when Sonnet 5 launched at the end of June. What’s become clearer since is that the rate increase isn’t the whole story.
Sonnet 5 shipped with a new tokenizer, and for a meaningful share of real workloads it produces around 30% more tokens than the previous model needed for the same input. That compounds with the price rise rather than sitting alongside it: a team whose Claude API bill assumed a 50% increase on 1 September could see something closer to a 90–95% increase once the token-count change is factored in, depending on how their specific code and prompts tokenize. There’s no extension announced this time — Anthropic has been generous with extensions on Claude Code’s usage boosts this year, but the Sonnet 5 pricing deadline has been fixed since launch and nothing in Anthropic’s recent communications suggests that’s changing.
Why this is worth checking now, not on September 1
Twenty-five days is enough time to actually do something about this, and almost no time at all if you wait until the invoice arrives. If your team or your development partner runs meaningful Claude Code or API volume, the useful move this week is a quick audit: pull August’s actual token consumption, apply both the rate change and a rough tokenizer-inflation estimate, and see what September’s bill looks like before it happens. That’s a very different conversation to have proactively than reactively.
So what
This doesn’t change the underlying calculus of whether Claude Sonnet 5 is the right model for a given job — it remains genuinely strong for agentic coding work, and Anthropic isn’t alone in restructuring pricing this year; OpenAI cut prices on its cheaper tiers just last week, in the opposite direction. What it does mean is that “AI usage costs” in a project budget deserve the same scrutiny as any other line item that’s about to move by double digits, not a fixed-cost assumption carried over from June. We factor model and token economics into project scoping rather than treating them as background overhead — see how that fits into AI-assisted development, or get in touch to talk through a build.