[AINews] GPT 5.6 price cut by 20%-80%: Cost of GPT 5.4 Intelligence dropped 13x in 4 months due to GPT 5.6 recursive self-optimization
GPT-5.6 recursively self-optimized OpenAI's own serving infrastructure, cutting costs 20-80%
“GPT-5.6 Sol was actively used to analyze production traffic, tune load balancing, and autonomously rewrite production kernels in OpenAI's own Triton and Gluon languages.”— Demis Hassabis
OpenAI's GPT-5.6 autonomously optimized its own inference stack—rewriting production kernels, tuning speculative decoding via hundreds of self-run experiments, and improving KV cache management—delivering a 20-80% price reduction. This marks a qualitative shift: a frontier model actively improving its own serving efficiency rather than relying solely on human engineers. Combined with the longer trend of GPT-4-level intelligence dropping 1000x in cost over 18 months, this suggests recursive self-optimization is now a meaningful accelerant in the AI cost curve.