OpenAI announced GPT-5.2-Codex on 18 December 2025, positioning the model as a successor to earlier Codex iterations with enhanced code generation capabilities [source]. The changelog describes improvements in multi-language support, context window expansion to 128,000 tokens, and claimed accuracy gains in complex algorithmic tasks.

The provider states GPT-5.2-Codex was trained on a dataset extending through mid-2025, including proprietary code repositories licensed from enterprise partners. OpenAI reports the model achieves 89.3% pass@1 on HumanEval, a benchmark measuring Python function completion, compared to 85.1% for the previous GPT-4.5-Codex release.

Notable changes include native support for Rust, Go, and TypeScript without explicit prompting, and a new "diff mode" that generates only modified code segments rather than full file rewrites. The model is available through OpenAI's API at a rate of $0.06 per 1,000 input tokens and $0.18 per 1,000 output tokens.

OpenAI's announcement does not address known issues from prior Codex versions, including documented cases of the model generating deprecated library calls or introducing subtle logic errors in edge-case handling. Independent developers reported similar behaviours in GPT-4-series models as recently as November 2025.

The provider has not published a technical paper or detailed evaluation methodology for the 89.3% HumanEval claim. OpenAI states the model will enter general availability on 2 January 2026, with early access granted to existing enterprise API customers beginning 20 December 2025.

This wire documents the provider's stated capabilities and release timeline. Observed performance in production environments has not yet been independently verified.