OpenAI has released the GPT-5.6 model family, which enhances price-performance by making frontier-level agent performance more affordable. The new models allow startups to achieve better results with lower costs through smarter model selection and improved API controls. According to OpenAI, the GPT-5.6 family continues the trend of tackling longer-horizon tasks with fewer tokens while maintaining strong agent performance and cost efficiency. The improvements in cost efficiency are accompanied by increased accuracy at lower reasoning efforts, as demonstrated in various production tests.
On the Agents’ Last Exam benchmark, GPT-5.6 Sol outperformed GPT-5.5 at higher reasoning levels when the harness was kept constant. Startups have also reported significant cost improvements across a range of workflows by reducing the reasoning effort from prior defaults. For instance, Izzy Miller, AI Research Lead at Hex, noted that using GPT-5.6 with low reasoning effort yielded the best results, as the model efficiently identified when data was insufficient and focused on reaching accurate conclusions with fewer tokens.
The GPT-5.6 family includes models like Luna and Terra, which offer high extraction accuracy at a fraction of the cost compared to previous models. Serhii Shchoholiev, Engineering Lead at Hypha, highlighted that Luna maintains 98% of GPT-5.5’s extraction accuracy at one-eighth the cost, making it practical for a wide range of workflows. Additionally, OpenAI has introduced new primitives to the Responses API to enable more efficient agent operations, including retained reasoning, parallel decomposition, and programmatic tool calling.
Source: openai
