Frontier models

GPT-6 Sol

GPT-6 Sol is OpenAI’s GPT-6 mid-tier model for complex coding and agentic workflows, released 22 September 2026 alongside GPT-6 Luna. It builds on GPT-6 Astra’s alignment work at lower API prices than GPT-5.6 Sol. The model id is gpt-6-sol. Pricing is $2 input and $10 output per million tokens (cached input $0.20; cache writes $2.50).

Status: Replaced, on the API; it still powers GPT-6 in ChatGPT for paid plans
AA score: 48 (max effort)Artificial Analysis Intelligence Index, one independent score across many tests, fetched 22 Sep 2026
Price: $2 in, $10 outUS dollars per million tokens (word pieces) in and out, unless noted
Released: 22 Sep 2026

Training

Knowledge cutoff on the Sol model page is April 20, 2026. Parameter count, training FLOPs, and the full train-time recipe are not published on the Sol docs page.

Inference

reasoning.effort supports none, low, medium (default), high, xhigh, and max. Reasoning tokens are supported. Use the Responses API for built-in tools and function calling. Chat Completions supports function calling only with reasoning_effort set to none.

Context

Documented context window is 1,050,000 tokens. Max output is 128,000 tokens. Text and image input; text output.

Methods

Named tools on the Responses API for this model include web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search. Streaming and structured outputs are supported. Fine-tuning is not supported on the model page.

Benchmarking

OpenAI’s 22 Sep community announce and model pages emphasize cost and availability over a full new harness table for Sol. Prefer Astra announce columns for older GPT-5.6 Sol comparisons, and prefer Artificial Analysis for independent Index.

Independent Artificial Analysis (22 Sep 2026 article): GPT-6 Sol (max) scores 48 on the Intelligence Index, about level with GPT-5.6 Sol, at roughly half the cost per Index task. Coding Agent Index in Codex harness: Sol (max) 57 (+2 vs GPT-5.6 Sol). AA notes lower hallucination on AA-Omniscience and mixed regressions on GDPval-AA. Artificial Analysis now marks GPT-6 Sol as deprecated on its site. OpenAI has since released GPT-6.1 Sol for the API.

Hardware

OpenAI has not published training or serving silicon details for GPT-6 Sol on the model docs. Buyers purchase API tokens.

Access and pricing

	GPT-6 Sol	GPT-5.6 Sol (prior)
Input / MTok	$2	$4
Output / MTok	$10	$20
Cached input / MTok	$0.20	(prior promo differed)

Available in the API, ChatGPT Work, and Codex for Plus, Pro, Business, Enterprise, and Edu per OpenAI’s announce. Prompts over 272K input tokens are priced at 2× input/cache and 1.5× output for the full request. Fast mode is 2× applicable rates.

In ChatGPT, OpenAI’s 7 Oct post says paid plans now use a GPT-6 Sol tuned for everyday conversation, with Intelligent UI in the Chat tab. Artificial Analysis scores the API model, not the ChatGPT version.

Sibling GPT-6 Luna (gpt-6-luna): $0.10 / $0.50 per million input/output; same 1.05M / 128K shape.