I Can’t Keep Up With The Frontier Model Updates
OpenAI’s new model is a top-end senior engineer—and easy to talk to
AI analysis
Questions asked and answered
- How does GPT-5.5 compare with Opus 4.7 on the Senior Engineer Benchmark? GPT-5.5 with extra high reasoning reached 62.5 on its best run. Opus 4.7 at a similar reasoning level landed in the low 30s. The article says human senior engineers score in the high 80s and low 90s.
- What does GPT-5.5 cost, and when is it available? API pricing is $5 per million input tokens and $30 per million output tokens, higher than GPT-5.4's $2.50 and $15. The model launches first in ChatGPT and Codex, and the API comes later while OpenAI completes additional safety and security validation. OpenAI argues that better reasoning and fewer retries can lower the cost per completed task despite the higher per-token price.
- Why does the article say GPT-5.5 feels different from earlier OpenAI models? The article says GPT-5.5 is built on a new pre-train, a base model that differs from the one used in earlier GPT-5.x models, which used the same pre-train as GPT-5.4. The article states that a new pre-train can change a model's underlying character, in contrast to post-training changes alone.
- Where does GPT-5.5 fall short, according to the article? The article says GPT-5.5 can be bland, struggles with Ruby, and trails Opus 4.7 on PowerPoint presentations, spatial composition, and ambitious prototypes. It also says Opus 4.7 seems to write better plans and has a stronger eye for design and product details.
Things the article might get wrong
The headline benchmark result comes from a test the publisher created for this review, the Senior Engineer Benchmark. The article does not describe how many tasks were run, how many runs were averaged, how the grading worked, or how the human senior engineer scores were obtained. The phrase "best run" suggests that the reported figure is the top result of several attempts, which may not reflect typical performance. It is also unclear whether the reasoning settings were matched in the way the phrase "similar reasoning level" implies. The article itself notes that GPT-5.5 performed best when it executed a plan written by Opus 4.7, which complicates the claim that GPT-5.5 is the strongest model overall.
Several central claims are presented without sources or figures. The article says GPT-5.5 is "much faster" than Opus 4.7 in head-to-head tests, but gives no timing data. The claim that a new pre-train explains the model's behavior is an interpretation, and the article does not say whether OpenAI confirmed the reasons behind the change. The argument that higher per-token prices may lower the cost per task comes from OpenAI's own pitch and was not tested by the publisher in the article.
The article has interests that readers may weigh. It is published by Every, which sells subscriptions, runs AI training services, and carries a sponsor (Hapax) in the body text. Most of the quoted user reactions come from people affiliated with Every or its newsletters, and the selection process for those quotes is not explained. The article also favors a positive framing of GPT-5.5 in its opening and headline, while the body contains more mixed evidence. Readers may want to look for independent benchmarks or reviews before drawing conclusions about the model's overall standing.

Seriously. The pace of frontier model updates is insane. Most days I feel like I'm being punked.
According to Every, "It’s much faster than Opus 4.7, easier to collaborate with, better at writing than any OpenAI model we’ve used since GPT-4.5 and GPT-4o, and the strongest model we’ve tested on our new Senior Engineer Benchmark, which measures how well models can rewrite a slop-coded codebase the way a senior engineer would."