model-release
GPT-6 Sol Release Date: Out Sept 22, 2026 — OpenAI's Coding-Agent Model, Rollout by Channel
Jacey
September 20, 2026
Updated on September 23, 2026
7 min read
GPT-6 Sol was released on September 22, 2026. It is OpenAI's model "built to power complex coding and agentic workflows", priced at half the GPT-5.6 Sol list rate. OpenAI's API changelog lists GPT-6 Sol and GPT-6 Luna as released that day, available through the Responses and Chat Completions APIs with text and image input and text output. The official model reference documents the model ID
gpt-6-sol, a 1,050,000-token context window, 128,000 maximum output tokens and a knowledge cutoff of April 20, 2026.GPT-6 Sol is live on EvoLink since September 23, 2026 as
gpt-6-sol, on the same API key you already use. The GPT-6 Sol API page owns the current rates, request examples and supported features; the GPT-6 Sol vs GPT-5.6 Sol guide owns the upgrade decision. This article owns the date and the rollout by channel.GPT-6 Sol rollout status by channel
| Channel or milestone | Status checked September 23, 2026 | Evidence |
|---|---|---|
| Official release date | September 22, 2026 | OpenAI API changelog entry naming GPT-6 Sol and GPT-6 Luna |
| Public API documentation | Live: model ID gpt-6-sol, Responses and Chat Completions | Official model reference |
| Public API pricing | Published: Standard, long-context (above 272K input tokens), Batch, Flex and Fast rows | Official pricing page |
| Codex | Rolling out from September 22 for Plus, Pro, Business, Enterprise and Edu users; community reports show some accounts not yet enabled | OpenAI announcement; staged per account |
| ChatGPT | Rolling out from September 22 in ChatGPT Work for Plus, Pro, Business, Enterprise and Edu; not offered on Free or Go (only Luna is, in the desktop app) | OpenAI announcement |
| EvoLink access | Callable since September 23, 2026 as gpt-6-sol; current rates on the API page | EvoLink route activation |
These rows move on their own evidence. A model visible in one Codex account is not the public API, and one channel's status says nothing about another's. The EvoLink rate and request examples live on the product page, where they are maintained with the route.
What changed from the pre-release tracker
On September 20 this tracker recorded no GPT-6 Sol entry in the OpenAI catalog or pricing page. Two days later both exist. The claims graded in the earlier version resolve as follows:
- "Sol was staged in the API" — a community report preceded the release by about two weeks. It was a useful signal to check the official sources; it was not, and still is not, evidence of access on any given channel.
- "GPT-6 Sol is a renamed GPT-5.6 Sol" — no. OpenAI documents
gpt-5.6-solandgpt-6-solas separate models with different knowledge cutoffs. Thegpt-5.6alias continues to point at the 5.6 generation. - "Sol will inherit Astra's contract" — no. GPT-6 Astra allows tool calling on the Responses API only. GPT-6 Sol supports function calling on both endpoints, but on Chat Completions only with
reasoning_effortset tonone. Read the model page, not the family name.
What GPT-6 Sol is built for, in OpenAI's words
OpenAI's model page positions GPT-6 Sol with one sentence: "Built to power complex coding and agentic workflows." The launch announcement adds the claims below. They are vendor-reported; none has been reproduced on EvoLink.
| OpenAI's claim | Reported figure | What it does not prove |
|---|---|---|
| Price | Half the GPT-5.6 Sol list rate, attributed to caching and inference improvements | Your EvoLink rate, or your cost per accepted task |
| Coding | DeepSWE v1.1: 68.8% at max effort, within 1.1 points of Claude Fable 5's 69.9% at xhigh, at roughly 80% lower cost per task | Results on your repositories, tools and acceptance tests |
| Agent honesty | 1.3% deception rate in OpenAI's internal coding-deception test, down from 10.4% for GPT-5.6 Sol | Behaviour under your tool permissions and prompts |
| Caching | Higher default prompt-cache hit rates; changing reasoning effort or tools no longer invalidates earlier cached context; 90% discount on cached input reads | Cache hit rates on your own prompts and route |
Sources: OpenAI announcement and model page, read September 23, 2026; secondary coverage by VentureBeat and TechCrunch. The upgrade guide turns these claims into an acceptance set you can run.
Why the Sol name still needs a version check
OpenAI now documents three Sol models across two generations: GPT-5.6 Sol, GPT-6 Sol, and the
gpt-5.6 alias that resolves to 5.6 Sol. A dashboard label, a client alias or a search result that says only "Sol" can describe either generation.Before acting on a claim, check three things: the exact identifier in the request, the identity returned in the response, and the knowledge cutoff (April 20, 2026 for GPT-6 Sol). A copied table of GPT-5.6 Sol specifications is not a GPT-6 Sol contract, even though the context window and maximum output are the same size.
What should an existing Sol team do now?
The illustration represents evidence filtering, not a completed GPT-6 Sol benchmark.
The release date is settled and the model is callable on EvoLink; the practical question is what the migration costs on your workload. Build a small evaluation pack before moving production traffic:
- Freeze representative tasks. Save a repository revision, issue description and test command for a small set of recent jobs. Include a failed job rather than selecting only easy successes.
- Record the existing contract. Capture the provider, exact model identity, reasoning setting, permitted tools, context construction and retry limit for your GPT-5.6 Sol route.
- Decide the endpoint. If your agent calls tools with reasoning enabled, GPT-6 Sol on Chat Completions will not do it; plan the trial on the Responses API or with
reasoning_effort: none. - Save the cost baseline. Include failed attempts and tool use. Compare cost per accepted task, not list price per token; the upgrade guide has the worksheet.
GPT-6 Sol, GPT-6 Astra and GPT-6 Luna are all callable on EvoLink; results on one say nothing about the others. For the current cost baseline, see the GPT-5.6 tiers and the GPT collection.
How this article will change
EvoLink activated GPT-6 Sol on September 23, 2026, one day after the OpenAI release; the two dates are recorded separately. Prices and request examples stay on the product page, where they are maintained with the route. Keep your GPT-5.6 Sol route selectable until your own evaluation clears the upgrade, and check current GPT-6 Sol rates before you size the trial.
FAQ
When was GPT-6 Sol released?
September 22, 2026. OpenAI's API changelog lists GPT-6 Sol and GPT-6 Luna as released that day through the Responses and Chat Completions APIs.
What is the GPT-6 Sol model ID?
gpt-6-sol, which is also the snapshot ID in the official model reference. EvoLink uses the same ID.Is GPT-6 Sol another name for GPT-5.6 Sol?
No. They are separate documented models. GPT-5.6 Sol is
gpt-5.6-sol; GPT-6 Sol is gpt-6-sol with a knowledge cutoff of April 20, 2026.Is GPT-6 Sol available on EvoLink?
Yes. Since September 23, 2026 you can call
gpt-6-sol on EvoLink with your existing API key, on Chat Completions and Responses. Current rates are on the GPT-6 Sol API page.Does a Codex appearance mean the public API is ready?
OpenAI's announcement rolls GPT-6 Sol out in Codex and ChatGPT Work for Plus, Pro, Business, Enterprise and Edu from September 22; the public API is a separate channel and is live for this model. Community reports of accounts still shown "not supported" are consistent with a staged, per-account rollout; check the notice for your plan.
What should I prepare before moving traffic to GPT-6 Sol?
Representative tasks, the current request contract, the endpoint decision for tool calling and a cost-per-accepted-task baseline. The linked upgrade guide explains how to use them.
