GPT 5.2

GPT 5.2 was released today and is an iterative upgrade to GPT 5.1.

OpenAI reports that enterprise users save 40 to 60 minutes per day, and heavy users save more than 10 hours per week.

On GDPval (Knowledge work tasks), GPT 5.2 reaches 70.9 percent compared to 38.8 percent in the previous model. On ARC AGI 2, performance rises to 52.9 percent from 17.6 percent.

Despite limited changes in coding benchmarks, in practical terms, GPT-5.2 feels better in professional workflows and coding. Still, benchmarks are simply benchmarks.

This is similar to how Anthropic's Claude Opus 4.5 set a new standard, demonstrating that quality depends as much on the model's harness and tools (Claude Code vs. Codex) as on raw model size or "capability".