Anthropic makes Claude Sonnet 5's $2/$10 pricing permanent, cancels September price hike

If you built a September budget line for Claude Sonnet 5 API costs, you can delete it. Anthropic confirmed on 10 August 2026 that the $2 per million input tokens and $10 per million output tokens rate for Claude Sonnet 5 is now its permanent standard price. The scheduled increase to $3/$15 per million tokens, which was due to take effect on 1 September 2026, will not happen.

The Anthropic platform docs pricing page now states this directly: “The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur.” The footnote that warned of the step-up has been removed entirely.

What was originally planned

When Claude Sonnet 5 launched on 30 June 2026, Anthropic framed the $2/$10 rate as introductory pricing, explicitly time-limited through 31 August 2026. Standard pricing of $3 per million input tokens and $15 per million output tokens was always listed as the rate that would apply from 1 September onwards. That 50% across-the-board increase is now cancelled.

This is worth pausing on. Sonnet 5 was already 33% cheaper than its predecessor, Claude Sonnet 4.6, which sits at $3/$15. It is unusual for a newer generation model to undercut the one it replaces. Anthropic locking in the lower rate makes that reversal permanent rather than a temporary promotional position.

What this means for your costs

This is not a price cut. Nobody’s bill goes down from where it is now. What has changed is that the increase you were expecting is not coming.

For teams processing modest volumes, the difference is small in absolute terms. But agentic workloads, customer-support pipelines, and coding tools that operate at scale tell a different story. The cancelled output price increase alone, from $10 to $15 per million tokens, works out to an extra $200 per month for every billion output tokens processed, or $2,400 over a year. For a product running at meaningful scale, that compounds quickly across workloads.

If your finance or engineering team had already started modelling Q4 costs on the assumption that Sonnet 5 output would rise to $15 from September, you can revise those figures down. The rate stays at $10.

A note on the newer tokenizer

One technical detail worth keeping in mind: Claude Sonnet 5 and later models use a newer tokeniser that produces approximately 30% more tokens for the same text compared with earlier generations. The exact figure varies by content type and workload shape. So while the per-token price is lower, token counts for equivalent text are higher. When comparing costs against older Sonnet models in production, account for this before assuming a direct saving.

Prompt caching and batch processing still apply

The permanent rate is the baseline. Anthropic’s prompt caching can reduce costs by up to 90% for workloads where large context windows are reused across requests, and batch processing carries a 50% reduction. These features remain available on top of the now-permanent $2/$10 pricing.

Who this does not affect

This is strictly about API pricing. Claude’s Free, Pro, Max, Team, and Enterprise subscription plans are separate products and were not part of the scheduled increase. If you access Claude through one of those plans rather than directly through the API, nothing changes for you here.

The competitive backdrop

Anthropic operates in a market where pricing decisions do not happen in isolation. With OpenAI’s comparable models available at similar price points, holding a 50% increase on a flagship mid-tier model through a period of intense competition would have been a meaningful risk. Keeping Sonnet 5 at $2/$10 maintains its position as a viable default for agent deployments, where the economics of running hundreds of millions of tokens per month are sensitive to even modest per-token shifts.

Claude Sonnet 5 has become the workhorse model for agentic use cases: multi-step reasoning, tool use, browser and terminal interactions, and document analysis at a price point that makes production deployments viable for most organisations. Keeping that price stable removes one variable from the build-versus-cost calculation for teams deciding whether to expand their usage.

What to do now

If you have already communicated a September cost increase to your team or adjusted forecasts upward, update those numbers. The $2 input / $10 output rate is the rate going forward. No action is required on the API side; your existing integrations using claude-sonnet-5 continue without change.

The Anthropic pricing docs are the authoritative reference if you need to share a source with finance or procurement teams.