May 8, 2026

Smaller replies, bigger bill

GPT-5.5 Price Increase: What It Costs

AI fans are side-eyeing the new price jump and asking if shorter answers are really worth it

TLDR: GPT-5.5 costs much more than GPT-5.4, and even its shorter answers only softened the blow, leaving many users paying 49% to 92% more. In the comments, people argued over whether it’s secretly more efficient overall or just an expensive guessing game with fewer obvious quality gains.

OpenAI’s new GPT-5.5 has arrived with a big price hike—about double the listed price of GPT-5.4—and the community reaction is somewhere between sticker shock, spreadsheet warfare, and full-on trust issues. The numbers in OpenRouter’s analysis say real-world costs rose 49% to 92%, even after GPT-5.5 trimmed its replies for very long prompts. Translation for normal humans: yes, it talks less in some cases, but many users still end up paying a lot more.

And the comments? That’s where the popcorn starts. One camp says the analysis misses the real story: if the new model solves tasks in fewer back-and-forth turns, maybe the higher price is worth it. Another camp is not buying that argument at all, calling the situation a “cost lottery” and questioning whether anyone can predict what they’ll pay. Then there’s the moodiest hot take of all: have chatbots already hit a progress slowdown? That comment landed like a dramatic season-finale cliffhanger.

The funniest energy came from benchmark obsessives doing their own math and basically yelling, “My wallet has entered the chat.” One user claimed GPT-5.5 was about 3.5 times more expensive on their tests, which only poured fuel on the debate. The vibe is clear: people don’t just want “better AI.” They want better AI that doesn’t make every prompt feel like ordering the market-price special.

Key Points

  • The article says GPT-5.5 launched at double GPT-5.4’s listed token prices, with input tokens rising from $2.50/M to $5.00/M and output tokens from $15/M to $30/M.
  • Using a switcher cohort of users who moved from GPT-5.4 to GPT-5.5, the analysis found actual costs increased by 49% to 92% across prompt-length buckets.
  • GPT-5.5 generated 19% to 34% fewer completion tokens for prompts longer than 10K tokens, partially offsetting its higher price for long-context use cases.
  • For prompts under 10K tokens, GPT-5.5 did not reduce output length; completions were 7% longer under 2K tokens and 52% longer in the 2K–10K range.
  • The analysis used OpenRouter request logs, compared April 21–23, 2026 with April 25–28, 2026, and normalized results as average cost per million OpenRouter tokens.

Hottest takes

"feels like a prime source of efficiency gains" — jsnell
"quite the cost lottery" — coalhouse
"~3.5x more expensive to run my benchmarks" — XCSme
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.