OpenAI has released a powerful model with GPT 5.2, but it hasn’t quite managed to unseat Google at the top.
OpenAI’s GPT 5.2 (High) model has placed second behind Google’s Gemini 3 Pro on the Artificial Analysis Intelligence Index. Also, GPT 5.2 Pro is slower and more expensive than Gemini 3 Pro on Artificial Analysis’s indices. Artificial Analysis Intelligence Index v3.0 incorporates 10 benchmarks that test models on skills in math, science and coding. On the Intelligence Index, GPT 5.2 (high) scored 73, the same as Google Gemini 3 Pro. GPT 5.2 managed to go past Claude Opus 4.5, which had scored 70.
GPT 5.2 (high) though was slower than Gemini 3 Pro. Gemini 3 Pro had a speed score of 119 on the speed index, while GPT 5.2 had a score of 90.
GPT 5.2 was also more expensive than Gemini 3 Pro. Gemini 3 Pro had a price of $4.5 per million tokens, while GPT 5.2 had a price of $4.8 per million tokens.
GPT 5.2 (High) placed second on the coding benchmarks as well, tying with Gemini 3 Pro was a score of 62. The coding benchmark represents the average scores in the LiveCodeBench, SciCode, Terminal-Bench Hard benchmarks. Claude Opus 4.5 followed with a score of 60.
GPT 5.2 placed second on the Artificial Intelligence Agentic Index as well, this time trailing Claude Opus 4.5. Google’s Gemini 3 Pro was placed third on this index.
All this suggests that GPT 5.2 is a step up from GPT 5.1 in capabilities, but hasn’t quite managed to outshine Gemini 3 Pro. The success of Gemini 3 Pro, and the resulting rise in usage for Gemini had caused OpenAI to declare an internal ‘Code Red’ at the company, with resources being marshalled towards creating capable models to take on Google. They seem to have done well with GPT 5.2, but Google appears to be still ahead — it’s marginally better than GPT 5.2 in capabilities, while being both cheaper and faster. It remains to be seen how things shape up in the coming weeks, but for now, Google’s Gemini 3 Pro seems to have maintained its spot at the top of the models pile.