Overview
OpenAI announced its latest AI model family “GPT-5.2” on December 11th. The release follows CEO Sam Altman’s internal “code red” memo earlier this month, clearly positioning it as a response to Google’s Gemini 3.
Three Model Variants
GPT-5.2 is available in three versions:
Instant
- Optimized for fast responses
- Designed for routine tasks like information-seeking, writing, and translation
Thinking
- Specialized for complex structured work including coding, math, and planning
- Excels at step-by-step logical reasoning
Pro
- For difficult problems requiring maximum accuracy and reliability
- Used when higher-quality answers are worth the wait
Benchmark Results
OpenAI introduced a new benchmark called “GDPval” that measures professional knowledge work tasks across 44 occupations.
- GPT-5.2 Thinking: Achieves performance at or above human professionals on 70.9% of tasks
- Gemini 3 Pro: Scores 53.3% on the same benchmark
- Error rate reduced by 30% compared to previous generation
On spreadsheet modeling tasks performed by junior investment banking analysts, scores improved from 59.1% with GPT-5.1 to 68.4%, a 9.3% increase.
Intensifying Competition
At the release, OpenAI’s CEO of Applications Fidji Simo denied that GPT-5.2 was rushed to market in response to Google. “These upgrades have been in development for many months,” she stated.
Meanwhile, Anthropic’s Claude Opus 4.5 scores higher on coding benchmarks in SWE-Bench Verified, but OpenAI argues this test is narrower and less representative of real-world use.
Future Developments
Simo also revealed that ChatGPT’s “adult mode” is expected to debut in Q1 2026, pending improvements to the age prediction model.
GPT-5.2 is rolling out to paid users and is available via API for developers.