GPT-5.3-Codex Benchmark Results Showing Incremental Gains Across Coding Tasks

Description

A benchmark comparison table on a black background with white text showing performance metrics for three OpenAI models: GPT-5.3-Codex (xhigh), GPT-5.2-Codex (xhigh), and GPT-5.2 (xhigh). The benchmarks include: SWE-Bench Pro (Public) at 56.8%/56.4%/55.6%, Terminal-Bench 2.0 at 77.3%/64.0%/62.2%, OSWorld-Verified at 64.7%/38.2%/37.9%, GDPval (wins or ties) at 70.9%/-/70.9% (high), Cybersecurity Capture The Flag Challenges at 77.6%/67.4%/67.7%, and SWE-lancer IC Diamond at 81.4%/76.0%/74.6%. GPT-5.3-Codex consistently leads across all benchmarks, with particularly large improvements on Terminal-Bench 2.0 and OSWorld-Verified