Reversal Confirmed in Official Documentation

Anthropic confirmed, in a note published on its official pricing documentation, that Claude Sonnet 5's promotional launch rate will not rise after all. The note states that the $2 per million input token and $10 per million output token pricing, introduced at launch as a temporary offer valid through August 31, 2026, is now the permanent standard price. The scheduled 50 percent increase to $3 and $15 per million tokens, set for September 1, simply will not happen.

What Was Originally Planned

When it launched Sonnet 5, Anthropic set the $2 and $10 per million token rate as a temporary introductory price, valid through the end of August 2026. Starting September 1, the price was due to rise to the same $3 input and $15 output per million tokens already charged for the previous model, Sonnet 4.6, a 50 percent increase on both sides. Outlets including Datafloq and Enterprise DNA had already been covering the deadline as a meaningful countdown for teams budgeting AI spend, recommending companies audit their usage before the increase landed.

Anthropic Chooses to Make the Discount Permanent

Rather than letting the increase take effect, Anthropic chose to make the promotional rate permanent instead. The company did not publicly detail its exact reasoning in the pricing note, but the move comes amid intense competition for market share among AI labs, with Google pushing Gemini 3.7 Flash as a cheap option for code and agents, and xAI expanding Grok's presence in developer tools such as GitHub Copilot. Keeping Sonnet 5 cheaper helps Anthropic sustain the model's appeal against that competition.

The Catch: a New Tokenizer Consumes More Tokens

Anthropic's own documentation carries an important caveat alongside the good news on frozen pricing: starting with Claude Opus 4.7 and later models, including Sonnet 5, a new tokenizer is in use, and it produces roughly 30 percent more tokens for the same text compared to the previous tokenizer used by Sonnet 4.6 and earlier models. In practice, that means even with the per token price held at $2 and $10, the real cost of processing the same task can end up higher than it was on the previous model, depending on the type of content and workload.

What This Changes for AI Budgets Through the Rest of 2026

For anyone who had already planned their AI infrastructure budget around the September increase, the cancellation is straightforwardly good news: projected spend for the coming months drops automatically, with no action required. Even so, the tokenizer caveat is a reminder that comparing per million token prices across model versions does not tell the whole story. Two model generations with the same headline price can produce different final bills depending on how many tokens each one consumes to deliver the same result.

Why It Matters for Brazilian Agencies and SMBs

For marketing and customer service automation agencies running high volumes of API calls against Claude, the news has an immediate, direct impact on cash flow: a September budget that already factored in a 50 percent cost jump can be revised downward right now. The recommended practical step is to rerun, with real usage data, a token consumption comparison between Sonnet 5 and the previous generation before finalizing contracts or passing prices on to end clients, since the new tokenizer can eat into part of the savings brought by the frozen price. It is also worth checking Anthropic's official pricing page regularly, since decisions like this one, favorable or not, tend to be announced there before any press coverage.