OpenAI: GPT-6 Sol
About this model
GPT-6 Sol is a closed-weight reasoning model from OpenAI, released on September 22, 2026, as a cost-efficient counterpart in the GPT-6 family alongside GPT-6 Luna and the flagship GPT-6 Astra. It is positioned for complex coding, agentic workflows, and computer-use tasks, with API access via gpt-6-sol and configurable reasoning effort from none through max. Pricing was cut by roughly 50% versus GPT-5.6 Sol, to $2 per million input tokens and $10 per million output tokens, with strong prompt caching for long agent runs.
On OpenAI’s launch evaluations, GPT-6 Sol at max effort reaches 68.8% on DeepSWE v1.1 and 56.4% on Agents’ Last Exam V1; at xhigh effort it scores 33.2% on AutomationBench 1.0.6 and 60.5% partial reward on OSWorld 2.0 offline. Independent runs at max effort report strong knowledge and agent scores, including about 94% on GPQA Diamond, 89% on MMLU-Pro, 80% on SWE-bench Verified, and mid-80s on Terminal-Bench 2.1, with a LiveBench composite near 79% and SimpleBench at 73.1%.
The model supports a 1.05M-token context window, up to 128K tokens of output, and text plus image inputs through the Responses API, with built-in tools such as web search, code interpreter, computer use, and function calling. It is available in ChatGPT Work, Codex, and the OpenAI API for paid tiers, emphasizing professional automation, software engineering agents, and sustained coding sessions at lower cost than frontier peers.
Benchmark Scores
Technical Specs
- Architecture: Transformer
- Context Window: 1,050,000 tokens
- Input Modalities: text, image
Hardware Requirements
- Compute: API only
Pricing
| Input | Output | Currency |
|---|---|---|
| 2.00 / 1M tokens | 10.00 / 1M tokens | USD |