OpenAI: GPT-6 Sol Pro (batch)

OpenAI: GPT-6 Sol Pro (batch) is the batch-inference variant of GPT-6 Sol served with reasoning.mode set to pro. It uses the same GPT-6 Sol weights and API surface as the standard model; pro mode allocates more internal reasoning work before returning a single final answer, which improves results on hard coding, analysis, and agentic workflows at the cost of higher latency and token usage. The batch endpoint is intended for high-volume, latency-tolerant workloads such as offline evaluation, enrichment, and large backfills, with the same per-token pricing as interactive Sol calls but asynchronous job semantics.

Released on September 22, 2026 alongside GPT-6 Luna, Sol targets complex software engineering and tool-using agents while cutting standard API list prices versus GPT-5.6 Sol. Official specs include a 1.05M-token context window, up to 128K tokens of output, text and image inputs, text outputs, and configurable reasoning effort from low through max. Knowledge is current through April 20, 2026. The model supports function calling, structured outputs, retrieval, web search, streaming, and prompt caching through the OpenAI API.

Reported benchmarks span broad knowledge (MMLU, MMLU-Pro, GPQA-Diamond), multimodal reasoning (MMMU-Pro), live contamination-resistant evaluation (LiveBench), factual QA (SimpleQA, SimpleBench), mathematics and code (MATH, HumanEval), repository repair (DeepSWE, SWE-Bench family), terminal and automation agents (Terminal-Bench 2.1, AutomationBench, Agents' Last Exam), web research (BrowseComp), and creative writing (Elo-style Creative Writing). Pro mode and max reasoning effort generally achieve the strongest published scores; batch serving does not change model behavior, only delivery timing.