Anthropic Quietly Raises Prices on Claude Sonnet 5 While Keeping Token Rates Unchanged
Anthropic’s latest model, Claude Sonnet 5, carries a hidden price increase that is masked by the same advertised token rates. The company changed the way tokens are counted, effectively making each request more expensive for developers and businesses.
The new model uses a different tokenization scheme that results in more tokens per prompt. That means customers pay more for the same amount of text, even though the price per token has not changed.
What Changed Inside Claude Sonnet 5
Tokenization now inflates costs. Claude Sonnet 5 uses a modified tokenizer that breaks text into smaller pieces. A sentence that previously consumed 10 tokens now takes 12 or more.
Pricing per token is static. Anthropic lists the same rates as before, such as $3 per million input tokens. But because more tokens are needed for identical input, the effective cost per request rises.
No public announcement. Anthropic did not notify customers in advance. The change was discovered by developers who noticed higher bills after migrating to Sonnet 5.
“We saw a 20% jump in token count for the same prompts. That’s a direct cost increase, but the official rate card never changed.” – Developer quoted in forum discussions.
How the Hidden Increase Impacts Users
Businesses face unpredictable costs. Budgets set for Claude 4 or Sonnet 4 now fall short. Teams that auto-scroll or use large context windows are hit hardest.
Developers lose comparability. Benchmarking between models becomes misleading because token counts no longer align. A “cheaper” model can be more expensive in practice.
Competitors may follow suit. If Anthropic normalizes this practice, other AI providers could adopt similar tokenization changes without transparent pricing.
What Anthropic Says – and Doesn’t Say
The company has not issued a formal statement. In support channels, representatives explain that the new tokenizer is “more efficient for the model” but do not address the cost impact on users.
No refunds or credits. Several customers reported that Anthropic refused to adjust billing because the per-token rate remains unchanged.
Documentation is vague. The tokenizer update is mentioned only in technical release notes. Most users learn about it only after seeing higher bills.
How to Protect Your Budget
Monitor token usage per request. Compare your average token count for the same prompt across model versions. Use an API wrapper to log actual consumption.
Set hard spending caps. Use Anthropic’s account limits to stop runaway costs. Flag any spike in token count for review.
Consider alternative models. If Claude Sonnet 5 inflates costs, test other providers or older models whose tokenizers stayed stable.
Demand transparency. Ask your account manager or support team for a clear breakdown of tokenizer changes and their effect on effective pricing.
The Broader Pattern
Anthropic has used similar tactics before. Previous model updates introduced new tokenizers without lowering per-token prices, effectively raising rates. The company relies on the complexity of tokenization to obscure real cost changes.
Customers who do not audit their usage will pay 10-25% more than they expect. The only warning sign is a higher bill.
Regulators have not yet addressed this practice. Industry observers say it exploits a gap in pricing transparency for AI APIs.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.