AI Agent Hub
Back to models
OpenAI: GPT-6 Sol Pro (batch) logo

OpenAI: GPT-6 Sol Pro (batch)

Closed Source openai Released 2026-09-22
48.0 / 100 1.1M context Proprietary

About this model

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.

Benchmark Scores

HLE
49.2
MATH
99.0
MMLU
96.5
DeepSWE
68.8
MMLU-Pro
88.9
MMMU-Pro
83.0
SimpleQA
60.7
HumanEval
97.0
LiveBench
79.25
SWE-Bench
74.0
BrowseComp
89.4
SimpleBench
73.1
GPQA-Diamond
96.0
AutomationBench
33.2
Agents-Last-Exam
32.2
Creative-Writing
2124.7
SWE-Bench-Verified
80.3
Terminal-Bench-2.1
83.15
SWE-Bench-Multilingual
81.7

Technical Specs

  • Architecture: Transformer
  • Context Window: 1,050,000 tokens
  • Input Modalities: text, image

Hardware Requirements

  • Compute: API only

Pricing

Input Output Currency
1.00 / 1M tokens 5.00 / 1M tokens USD