What is GPT-6.1 Sol?
GPT-6.1 Sol is OpenAI's mid-tier model in the GPT-6 family, released on September 29, 2026 as an upgrade to GPT-6 Sol, which had shipped a week earlier. OpenAI's announcement pitches it as "near-Astra intelligence for a fifth of the price": $2 per million input tokens, $10 per million output tokens and $0.10 for cached input. Its model page lists the same 1,050,000-token context window and 128,000-token output limit as Astra, plus an Ultrafast mode at six times the standard price.
We ran it in Codex at extra-high reasoning on eight briefs, one attempt each, with no human edits, exactly as we ran every other model.
What did GPT-6.1 Sol build?
Eight complete projects, for about $6.40 in total at list prices.
| Brief | Finished | Wall time | Output tokens | Files | Source size | Runtime errors | API-equivalent cost | Attempts |
|---|---|---|---|---|---|---|---|---|
| SaaS landing page | Yes | 9m 08s | 45k | 1 | 73 KB | 0 | $0.69 | 1 |
| Analytics dashboard | Yes | 7m 26s | 36k | 1 | 67 KB | 0 | $0.54 | 1 |
| Balls in a spinning hexagon | Yes | 7m 54s | 43k | 1 | 25 KB | 0 | $0.65 | 1 |
| First-person shooter | Yes | 13m 19s | 61k | 3 | 65 KB | 0 | $1.03 | 1 |
| Mobile finance app | Yes | 13m 46s | 48k | 1 | 57 KB | 0 | $0.83 | 1 |
| One-lane MOBA | Yes | 19m 41s | 67k | 3 | 85 KB | 0 | $1.12 | 1 |
| Open brief | Yes | 10m 51s | 40k | 1 | 65 KB | 0 | $0.71 | 1 |
| Kanban board | Yes | 14m 31s | 53k | 1 | 64 KB | 0 | $0.83 | 1 |
- Landing page. "Less busy. More you." in DM Sans and Instrument Serif, a terracotta accent and an interactive booking dialog. Nine minutes, $0.69.
- Analytics dashboard. "Room to grow." in DM Serif Display, with a deep green sidebar and correct figures throughout. Seven and a half minutes, $0.54, the cheapest build on the site.
- Physics simulation. Kinetic again, "Chaos, contained.", on a light background with a trails toggle and every ball still inside at the end. Eight minutes.
- 3D shooter. Dead Link: a green-tinted yard with a sign reading Relay 09. The player dies 35 seconds into our 40-second script. Thirteen minutes, three files.
- Phone finance app. "Hey, Jordan." under a drawn iPhone status bar, a "Step 1 of 4" send flow and a slide-up review sheet; the balance updates to $4,783.69 after sending $42.50. Fourteen minutes.
- One-lane MOBA. Aetherfall, a dark, sparse arena whose hero, Kael, casts Rift Lance, Aegis Field, Phase Step and Starfall. The hero falls once during our 60-second script. Twenty minutes.
- Open brief. It invented afterhours, a $59 physical off switch for your phone life, under "LESS SCROLL. MORE SOUL." Eleven minutes.
- Kanban board. An orange-accented board with a team sidebar and a "Good things are growing." card, with working drag and drop and undo. Fourteen and a half minutes.
Does "near-Astra" show in the builds?
Yes, to an almost uncanny degree. Put Sol's builds next to Astra's and you see the same decisions: headlines built on "Less busy. More…", the same pair of typefaces on the landing page, a physics demo with the same name, MOBA heroes that both end on an ultimate called Starfall (a name Claude Fable 5.1 also chose), kanban and finance layouts that differ mainly in accent colour. The public Terminal-Bench 4.0 leaderboard agrees from a different angle: the two tie at 58.2%, and Sol's run cost about $600 against Astra's $3,300.
Where they differ, Sol was faster on six of the eight briefs, Astra on the finance app and the kanban board. And in the shooter, Astra's player survived our script while Sol's did not.
How did GPT-6.1 Sol work?
The same way as Astra, with slightly more checking. Codex ran each Sol build as a single agent turn of 14 to 33 shell commands. Across all eight builds Sol read about 12.7 million input tokens, 95% of them from the prompt cache at $0.10 per million, and wrote about 394,000 output tokens including reasoning, a little more than Astra wrote for the same briefs.
That is the economics behind Sol's price per build. With almost all input cached at a twentieth of the normal rate, a Sol build's bill is mostly its output tokens at $10 per million, which is why even its largest build, the MOBA, came to $1.12.
What is GPT-6.1 Sol good at?
Turning a clear brief into a working build, cheaply. Every Sol build includes every feature the prompt listed, in a single agent turn, and the source stays compact and readable: 25 to 85 KB per build. That compactness is a feature in its own right: a developer can read a whole Sol build in an afternoon and change it with confidence. On the data-heavy briefs it computed the right numbers and kept them consistent after interactions. Its accessibility markup is, if anything, more thorough than its siblings': its landing page carries more ARIA attributes than any of the other three.
Where does GPT-6.1 Sol fall short?
Ambition and robustness under pressure. Sol builds exactly what was asked and rarely more, in the same house style as Astra. In the two games, its arenas are the plainest of the four and both its heroes fall during our scripted input. On the public leaderboard it trails Claude Opus 5.5 by about six points. None of that matters for a prototype; all of it matters for the build you hand to a client.
How does GPT-6.1 Sol compare with Claude?
As a fast first draft against a finished one. Claude Opus 5.5 and Claude Fable 5.1 took 25 minutes to two and a half hours per build on our briefs and cost $7 to $21 each at list prices; Sol took 7 to 20 minutes and cost about a dollar. The Claude builds are richer and more varied. For many projects the right answer is both: draft with Sol, then decide whether the job deserves Claude's hour. Opus 5.5 against Sol shows the gap on every brief.
What would GPT-6.1 Sol cost you?
Very little, which changes how you can work. All eight of our builds together came to about $6.40 at list API prices, less than a single build by any Claude model. At that price it is reasonable to generate several versions of a page or tool and keep the best one, which is the most reliable way around Sol's tendency to pick the safe default.
OpenAI also lists cheaper and faster ways to run it on its model page: Batch and Flex requests at half price for work that can wait, and Ultrafast mode at six times the standard price when speed matters more than cost. Prompts over 272,000 input tokens are billed at twice the input rate, which long agent sessions can reach.
Who should use GPT-6.1 Sol?
Anyone who builds often and iterates. Startups prototyping, developers scaffolding internal tools, designers exploring directions and anyone paying per token will get more working builds per dollar from Sol than from any other model we test. Choose something else only when the first attempt has to be the final one. And if you do use Sol for client work, budget an hour of your own for the personality it will not add by itself: a distinctive typeface, a bolder colour, one surprising interaction. The structure underneath will already be sound.
Frequently asked questions
What is GPT-6.1 Sol?
An upgrade to GPT-6 Sol released by OpenAI on September 29, 2026. OpenAI says it nearly matches GPT-6 Astra on agentic coding, computer use and professional work at one fifth of Astra's token prices: $2 per million input tokens and $10 per million output tokens.
Is GPT-6.1 Sol as good as GPT-6 Astra?
For building, essentially yes. It ties Astra on the public Terminal-Bench 4.0 leaderboard at 58.2%, and on our eight briefs its builds were strikingly similar to Astra's. Its one clear loss was the 3D shooter, where its player died before our scripted input ended.
How much did GPT-6.1 Sol's builds cost?
Between $0.54 and $1.12 each at list API prices, about $6.40 for all eight. The same eight briefs cost about $31 with GPT-6 Astra and $7 to $21 per build with the Claude models.
Is GPT-6.1 Sol available in ChatGPT?
OpenAI lists GPT-6.1 Sol on its Business plan and includes an Ultrafast version in the $500 Pro tier. For Plus, OpenAI's help page advises checking the model picker for current access. It is also available on the API and in Codex.