How GPT-5.6 fuses frontier intelligence with frontier efficiency

OpenAI announced Tuesday the GPT-5.6 model family, designed to balance capability and cost across a range of tasks. The company said its flagship model, GPT-5.6 Sol, outperforms Claude Fable 5 on the Artificial Analysis Coding Agent Index at less than half the cost. OpenAI said the Terra model performs as well as GPT-5.5 on intelligence benchmarks at half the price, while Luna is priced 80% less than Sol. The company said it achieved its greatest intelligence-per-token efficiency by training models to take a more direct path through tasks. OpenAI said optimizations span models, inference, and the agentic harness used by Codex and ChatGPT Work. The company said GPT-5.6 Sol in Codex played an instrumental role in optimizing the inference stack, including load balancing, kernel optimization, and speculative decoding. OpenAI said load balancing improvements alone dramatically reduced serving costs. The company said GPT-5.6 Sol autonomously rewrote and optimized production kernels in Triton and Gluon, reducing end-to-end serving costs by 20%. OpenAI said speculative decoding improvements increased token-generation efficiency by more than 15%. The company said it used GPT-5.6 Sol to hyper-optimize engine and model configurations for each scenario. OpenAI said its mission is to ensure artificial general intelligence benefits all of humanity.