GPT-5.6 Sol scored 72.7% on DeepSWE benchmark

GPT-5.6 Sol, Terra, and Luna launched on Amazon Bedrock

Open-source MoE model matches GPT-5.5 on agentic coding benchmarks

GPT-5.5 achieves 16% Pass@8 on MA-ProofBench Level I and 5% on Level II, outperforming most models.