tokevenAre AI API Tokens Getting More Expensive?
We tracked every pricing change across Claude, GPT, and Gemini APIs to find out. The answer depends on the model tier you use. Flagship models occasionally rise, but competition drives mid-tier and budget prices steadily down.
The data
Pricing change summary
Total changes
12
Price cuts
5
Price increases
3
Avg repricing change
+273.5%
The trend is mostly down
Of the 12 pricing events tracked, 5 were price reductions or competitive new-model launches, while only 3 represented an increase. New model launches almost always undercut or match the previous generation on a per-token basis, and when they do cost more, the price increase reflects a meaningful capability jump.
The biggest savings come not from waiting for price cuts, but from choosing the right model for each task. A flagship model running classification work can cost 10-20x more than a budget model that handles the task just as well.
Provider-by-provider breakdown
What matters more than list price
List-price changes capture only part of the cost picture for your team. Three factors matter more for your total bill in production.
Caching
Cache-read discounts of 75 to 90 percent off input costs are available across all three providers. If your prompts repeat system messages or context, caching alone can cut your bill in half.
Model selection
Budget models handle classification and extraction at confidence levels above 80 for far less. Matching the right model to each task is the single biggest cost lever you have.
Prompt efficiency
Shorter, focused prompts with clear instructions reduce token counts while preserving output quality. A 30% reduction in prompt length saves 30% on input costs for every model.
Full data
All pricing changes
Frequently asked questions
Why do AI tokens cost more?
AI token prices depend on GPU compute costs, model parameter count, and market competition. Larger models with more parameters cost more to run per token. Competition between Anthropic, OpenAI, and Google has driven prices down over time for mid-tier and fast models. Price increases tend to appear when providers release upgraded models that demand more compute resources.
Which provider has the best price trend?
Across 12 tracked changes, 5 were price cuts and 3 were increases. Claude leads in total price reductions so far. All three major providers compete aggressively on budget and mid-tier pricing. Flagship models tend to hold or rise in price as capabilities improve.
How to protect against token inflation?
Three strategies work best. First, use prompt caching, where cache discounts range from 75% to 90% off input costs. Second, right-size your model, because budget models handle classification and extraction at 5-10x lower cost. Third, monitor per-call costs with Tokeven so you spot price changes and switch models early.
Explore
Related tools
Track your real costs
Pricing pages show list rates for each model. The dashboard shows your real spend per call, per model, and per team.