OpenAI: GPT-6.1 Sol Pro (batch)

GPT-6.1 Sol Pro (batch) is OpenAI’s high-reasoning configuration of GPT-6.1 Sol, exposed through the Batch API for asynchronous workloads at reduced per-token rates. It uses the same underlying weights as standard GPT-6.1 Sol but sets reasoning.mode to pro so the model spends more internal reasoning before returning a final answer. OpenAI positions this mode for difficult, high-stakes tasks—complex agentic coding, computer use, multi-tool business workflows, and dense document understanding—where extra accuracy justifies higher latency and token usage than the default Sol setting.

The model supports a 1.05M-token context window, up to 128K output tokens, text and image inputs, structured outputs, function calling, and the same agent tooling stack as GPT-6.1 Sol (web search, code interpreter, computer use, MCP, and related hosted tools). The batch endpoint is intended for offline or delay-tolerant jobs such as large-scale evaluation, enrichment, and backfill pipelines rather than interactive chat. Cached input pricing and batch discounts make long-context and high-volume pro-reasoning runs more economical than synchronous pro calls at list rates.

Reported evaluations for the GPT-6.1 Sol family emphasize real-world agent benchmarks—DeepSWE for software engineering, AutomationBench for multi-step office workflows, LiveBench for contamination-resistant broad skills, and multimodal suites such as MMMU-Pro—alongside coding edit benchmarks (SWE-Bench Verified, Aider) and graduate-level knowledge tests (MMLU-Pro, GPQA). Pro mode generally improves hardest-task scores relative to default Sol while remaining substantially cheaper than GPT-6 Astra for comparable agentic workloads.