OpenAI launches GPT-5.6 with Codex in ChatGPT

GPT-5.6 achieves SOTA on multiple benchmarks, including 7.8% on ARC-AGI-3 and 73% on DeepSWE. The Sol Ultra variant reportedly proved a 50-year-old math conjecture. The model is now the preferred option in Microsoft 365 Copilot and available via Perplexity's Agent API.