The cost of the GPT‑5.5 model from OpenAI exceeded that of the previous version by 1.5 to 2 times

New Prices for GPT‑5.5: What Changed and Why It Costs Money

With the release of the GPT‑5.5 model, OpenAI raised its access cost via API. In some cases, the price per token is twice as high as the previous version.

OpenAI claims that the price increase is partially offset by higher model efficiency: GPT‑5.5 delivers better results using fewer tokens.

How It Looks in Practice
OpenRouter analysis showed that the actual cost of working with GPT‑5.5 rose from 49 % to 92 % compared to GPT‑5.4:

- For long requests (over 10,000 tokens) savings come from reduced execution time, though a noticeable price increase remains.

- For short requests (< 10,000 tokens) the cost rises significantly, but processing time does not decrease.

Additionally, GPT‑5.5 generates 19–34 % fewer tokens on long requests, reducing system load but not fully negating the price hike.

Financial Consequences for Industry Leaders
OpenAI and Anthropic continue to operate at a loss:

- OpenAI: expected loss of $14 billion in 2026.
- Anthropic: projected loss of $11 billion.

Anthropic’s new model, Claude Opus 4.7, is sold under the same rates as its predecessor despite claimed tokenizer improvements. However, OpenRouter found that:

- For requests longer than 2,000 tokens, costs increased by 12–27 % even with caching.
- For requests shorter than 2,000 tokens, shorter responses helped offset additional tokenizer expenses.

Conclusion
Deploying GPT‑5.5 has led to a significant rise in API prices, but the company remains confident in the model’s enhanced efficiency. In practice this means noticeable price increases for users—especially with short requests—and continued financial losses for the largest AI market players.