OpenAI launched GPT-5.6 on July 9 in three capability and cost tiers. Its tables can frame a test, but cannot name a winner without reproducing configuration, quality, time, tokens, and price.
OpenAI trained GPT-5.2 and GPT-5.3 Codex entirely on NVIDIA infrastructure. The latter is its first agentic coding model to take part in its own development. Behind it lies a hardware race that now determines who gets to build frontier AI.
Johannes Heidecke is leaving OpenAI after its safety teams were placed under joint research and safety leadership. The organization chart matters, but it is not enough to measure control over risk.
OpenAI releases GPT-4.5, its largest and most knowledgeable model yet. It promises fewer hallucinations, while its tests show strengths distinct from reasoning models.
OpenAI measured Codex and published two figures: it solves 28.8% of problems on the first attempt and 70.2% given a hundred. Both are true and only one is usable, because choosing among a hundred answers requires knowing which one works. What Codex is per its own paper, how “can it code” gets measured, where it fails, and why it does not learn from you as you use it.
This website uses cookies to improve the browsing experience. Cookie policy.
No results. Try different words.
