Jul 30, 2026 13:30:00

OpenAI autonomously improved the inference efficiency of GPT-5.6 using GPT-5.6 itself.

OpenAI

released the GPT-5.6 series on July 9, 2026. OpenAI has stated that the GPT-5.6 series utilizes GPT-5.6 itself to reduce computational costs and improve GPU utilization efficiency.

How GPT-5.6 fuses frontier intelligence with frontier efficiency | OpenAI
https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/

After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run.

The result:
- 20% lower costs serving from production GPU kernel improvements.
- 15%+ better token-generation efficiency from improved speculative decoding.

— OpenAI (@OpenAI) July 29, 2026

The GPT-5.6 series was released in three variations: the high-performance 'GPT-5.6 Sol,' the balanced 'GPT-5.6 Terra,' and the low-cost 'GPT-5.6 Luna.' Of these, the GPT-5.6 Sol is marketed as a model that can achieve performance exceeding that of the Claude Fable 5 at half the cost.

OpenAI officially releases 'GPT-5.6,' demonstrating coding performance exceeding Claude Fable in third-party tests - GIGAZINE

According to OpenAI, the high cost-effectiveness of GPT-5.6 Sol is achieved through self-improvement using GPT-5.6 Sol itself.

In inference, the inference engine kernel itself was improved with GPT-5.6 Sol. GPT-5.6 Sol has learned syntax that contributes to the efficiency of ' Triton ' and ' Gluon, ' the programming languages used in OpenAI's kernel, and it is said that kernel optimization has succeeded in reducing end-to-end service costs by as much as 20%.

Improvements to speculative decoding were also implemented autonomously. Speculative decoding is an inference acceleration technique that uses a small 'draft model' to predict the next token, and token generation efficiency has been improved by 15% by improving the draft model using GPT-5.6 Sol.

Furthermore, the handling of the KV cache and the system for allocating computational processing to the GPU are also being improved autonomously.

Furthermore, OpenAI has also streamlined the agent system harness. Specifically, they have made improvements such as 'calling MCP servers and skills only when needed,' 'limiting the maximum number of tokens output by tools to 10,000 by default, and only allowing more tokens if requested by the model,' and 'fixing the order in which tools are called,' which have reportedly improved cache utilization.

OpenAI commented, 'We are confident that the pace of optimization will accelerate going forward, as GPT-5.6 has made a significant contribution to improving inference and harnesses.'