What is GPT-5.6?

GPT-5.6 is a game-changer - it's a significant upgrade in the GPT model family, and it's all about offering a new standard for price-performance. This means frontier-level agent performance is now more affordable, which opens up a whole world of possibilities. Startups, for instance, can build faster, more capable agents at a fraction of the cost, which is a huge deal. And the best part? The model's improvements in cost efficiency are compounded with increased accuracy at lower reasoning efforts, making it a no-brainer for a wide range of applications.

The GPT-5.6 model family has some exciting additions, like the smaller models Luna and Terra, which can perform similarly to flagship models like GPT-5.4 and 5.5, but at a significantly lower cost. These models are perfect for high-volume workloads, latency-sensitive interactions, and repeated steps within agentic workflows. Take a legal-tech startup, for example - they can use Terra or Luna for extraction and save a pretty penny. And with the new Responses API primitives, agents can operate more efficiently by reusing work already performed, decomposing tasks in parallel, and moving deterministic work into code.

The impact of GPT-5.6 is already being felt, with startups reporting some pretty substantial cost improvements across a range of workflows. On the BrowseComp benchmark, for instance, GPT-5.6 Luna delivers essentially the same performance as GPT-5.5, but at a fraction of the cost. And with the new Responses API capabilities, agents can finish complex tasks faster and more efficiently. As a result, GPT-5.6 is now the default model for many high-throughput code retrieval and decision modeling workloads - it's a combination of capability and cost that's pretty remarkable for browser agents.

Source: OpenAI