GPT-5.2 arrives to end OpenAI's code red — with a 40% input-price increase
OpenAI released GPT-5.2 on December 11 in Instant, Thinking and Pro tiers, roughly a month after GPT-5.1 and after CEO Sam Altman issued an internal code red memo redirecting resources to ChatGPT in response to Google's Gemini 3. The model carries a 400,000-token context window and an August 31, 2025 knowledge cutoff. OpenAI says GPT-5.2 Thinking beats or ties human professionals on 70.9% of comparisons on GDPval, its benchmark of knowledge-work tasks across 44 occupations, and that it sets a new state of the art on OpenAI's MRCRv2 long-context evaluation and at 98.7% on Tau2-bench Telecom for tool calling.
OpenAI's launch page lists no head-to-head numbers against Gemini 3; comparisons are against OpenAI's own predecessors, with competitive benchmarks against Gemini 3 Pro and Claude Opus 4.5 shown only during the press briefing. API input pricing rose to $1.75 per million tokens, about 40% above GPT-5.1, and Anthropic's Opus 4.5 still scores higher on SWE-bench Verified. Altman told CNBC that Gemini 3's impact on OpenAI's metrics was smaller than feared and that he expects to exit code red by January.
WHY IT MATTERS
The capability number went up and so did the price, so re-run your cost model rather than your latency model: push bulk, batch and long-context work to cheaper or open-weight models and reserve GPT-5.2 Thinking for hard reasoning. The missing Gemini 3 comparison is the one your model-selection decision actually needs — vendor-run benchmarks citing OpenAI's own prior models are not a substitute for your own eval set on your own tasks.
