ai and ml
It’s not just gas prices skyrocketing. Frontier-model pricing keeps climbing too
It’s getting more expensive to use the latest models. OpenAI last month bumped the version number of its GPT model family to 5.5, and per-token prices rose too, in some cases doubling compared to its predecessor.
For 1 million tokens, GPT-5.5 is priced at $5 (input), $0.50 (cached input), and $30 (output). Its predecessor GPT-5.4 charges $2.50 (input), $0.25 (cached input), and $15 (output) per 1 million tokens.
The AI biz claims that the cost increase is offset to some extent by token processing efficiency – delivering better results using fewer tokens.
“While GPT‑5.5 is priced higher than GPT‑5.4, it is both more intelligent and much more token efficient,” the company said during the rollout.
But the cost is still going up, more than efficiency improvements are reducing costs. According to an analysis conducted by OpenRouter, GPT-5.5 is anywhere from 50 percent more expensive to nearly twice as expensive, depending on prompt length.
“Our analysis shows that GPT-5.5 actual costs increased 49 percent to 92 percent,” OpenRouter said. “Longer prompts, over 10k tokens, saw costs offset by shorter completions. Shorter prompts, under 10k, experience a higher cost increase where completions did not get shorter.”
That range – 49 percent to 92 percent – factors in the model’s token efficiency improvements, which are more relevant for longer prompts. According to OpenRouter’s measurements, GPT-5.5 generates between 19 percent and 34 percent fewer completion tokens for longer prompts (10,000 tokens and up).
If reports of OpenAI’s projected $14 billion loss in 2026 prove accurate, costs will have to rise much more to balance its insistent spending. But this is a problem also faced by rival Anthropic, set to lose a reported $11 billion in 2026.
Anthropic’s Claude Opus 4.7 arrived without a visible list price change amid claims about an improved tokenizer. The result, according to OpenRouter, is potential savings for shorter prompts but larger bills for longer ones.
“Our study of real Opus 4.7 usage shows that actual costs increased 12–27 percent for prompts above 2K tokens when cache absorption is taken into account,” the biz said. “Short prompts under 2K were the exception, where significantly shorter completions offset the tokenizer overhead entirely.”
Expect further price increases for premium models. ®
