GPT 6.1 Terra vs Sol comes down to this: pick Terra for fast, high-volume computer use, screen reading, forms and writing, and pick Sol for hard coding and long, complex jobs where a few extra points of accuracy matter more than speed. Terra costs half as much per token, and OpenAI says it finishes tasks over three times faster than Sol.
📺 Watch: GPT-6.1 Terra Is INSANE!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
OpenAI released GPT-6.1 Terra on 26 October 2026, about four weeks after GPT-6.1 Sol landed on 29 September. Both sit in the same family under GPT-6 Astra, with GPT-6 Luna at the cheap end. Here's what each one is for, the numbers side by side, and which one I'd plug into my agents.
What each model is built for
OpenAI calls Terra a fast, efficient model that combines detailed visual reasoning, precise computer use and clearer, more natural writing. Its examples are everyday work: drafting proposals, reading technical drawings, filling out forms and editing videos. Microsoft's Foundry launch post frames it the same way. Terra handles the high-volume visual and screen-based tasks, while Astra plans the bigger job.
Sol is the step up. OpenAI's model page describes it as near-Astra performance for complex work at a lower cost, and the ChatGPT model guide recommends it for repeated, long-running work across code, apps and documents.
So think of it like a team. Terra is the quick operator that clicks, reads and writes all day. Sol is the senior engineer you hand the tricky stuff. If you want to see how I wire both into one system, the AI Profit Boardroom ($69/mo) gives you my Agent OS, a 30-day AI agent roadmap, daily tutorials as new models drop, 4 coaching calls a week and the prompt library. 3,400+ members are already in.
GPT 6.1 Terra vs Sol side by side
Every number below comes from OpenAI's announcements and model pages unless marked Microsoft. Benchmark scores are at max reasoning effort.
| GPT-6.1 Terra | GPT-6.1 Sol | |
|---|---|---|
| API model ID | gpt-6.1-terra | gpt-6.1-sol |
| Context window | 1,050,000 tokens (922K input, 128K output) | 1,050,000 tokens (922K input, 128K output) |
| Knowledge cutoff | 1 June 2026 | 30 April 2026 |
| API price per 1M tokens | $1 input, $0.05 cached, $5 output | $2 input, $0.10 cached, $10 output |
| Speed | Over 3x faster than Sol on Agents' Last Exam tasks; tokens per second not published | Ultrafast mode listed at 6x the standard price; tokens per second not published |
| Reasoning effort | none to max | low to max (no none) |
| Agents' Last Exam v1 | 60.0% | 58.6% |
| OSWorld 2.1 (computer use) | 74.2% | 78.0% |
| MMMU-Pro (vision) | 85.5% | 86.3% |
| Terminal-Bench 4.0 (coding) | 50.3% | 55.9% |
| Computer use tool in API | Yes | Yes |
| ChatGPT | Plus, Pro, Enterprise, Edu in ChatGPT Work and Codex | Plus, Pro, Business, Enterprise, Edu in ChatGPT Work and Codex; not yet in Chat |
| Microsoft Foundry | Generally available, $1 input and $5 output on Global Standard (Microsoft) | Not covered in this comparison |
Two things jump out. The context window is identical, so you're not giving anything up on long documents. And the gap on vision and computer use is small, while the price gap is a clean 2x.
📺 Watch: Qwen 3.8-Max, Fable 5 and GPT-5.6 Sol: Who Wins?
Coding: Sol still wins
OpenAI is straight about this one. Terra is a big jump over GPT-5.6 Terra on Terminal-Bench 4.0, going from 21.5% to 50.3% at max effort. But OpenAI also says that for the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice.
The chart shows why. Sol at medium effort scores 52.7% for about $1.02 per task. Terra needs max effort to reach 50.3%, and that costs about $1.84 per task. So on coding, the cheaper token price doesn't make Terra the cheaper run. Sol gets a better score for less.
There's one exception. On Terminal-Bench-Science 0.1, Terra at max scores 58.3% at $4.15 per task, and Sol at max scores 57.0% at $5.47. For code-heavy research work, Terra edges it.
My take: keep Sol as your coding model. For the deep dive on that, read my guide to GPT-6.1 Sol for coding. If you want a hand picking the right model for each job in your business, grab a free AI strategy session.
📺 Watch: GPT-6 + Doug: OpenAI's Secret AI
Agents and computer use: Terra's home turf
This is where Terra earns its place. On Agents' Last Exam, a set of professional tasks like turning architectural drawings into a 3D model, Terra at medium effort scores 57.6% at about $0.29 per task. Sol at medium scores 53.8% at about $1.31. Terra is higher and roughly a fifth of the cost. OpenAI also says Terra matches GPT-6 Astra there at one-tenth the price per token.
On OSWorld 2.1, which tests desktop apps over many steps, Sol keeps a lead at every effort level. At medium it's 73.5% for Sol against 66.7% for Terra, and at max it's 78.0% against 74.2%. So for the longest, messiest desktop workflows, Sol is still more reliable.
On factual errors, they're close. At max effort, Terra's error rate on OpenAI's hard prompts is 4.3% and Sol's is 4.4%. At low effort, Sol is a bit better at 14.7% against Terra's 16.7%.
In my video on Terra, I set up three Terra agents: a Search Console report agent, a lead agent and an admin agent. Those are perfect Terra jobs. They read screens, click through the same steps every day and write a clean summary. In my Agent OS, Terra is the everyday model and Sol or Astra stays ready for the hard jobs. The full setup is in my Agent OS guide.
Cost math with OpenAI's prices
Using OpenAI's standard API prices, here's a simple example. Say an agent reads 1,000 invoices, each one using 20,000 input tokens and 2,000 output tokens. That's 20 million input and 2 million output.
- Terra: 20M at $1 is $20, plus 2M at $5 is $10. Total: $30.
- Sol: 20M at $2 is $40, plus 2M at $10 is $20. Total: $60.
Cached input follows the same 2x pattern: $0.05 per million for Terra against $0.10 for Sol. Both models also get Batch and Flex at 50% off standard, and both charge 2x input and 1.5x output on any request over 272K input tokens.
But token price isn't the whole story. Effort level changes the bill per task. As the coding numbers showed, a model that needs max effort can end up costing more than a pricier model at medium. Test both on your real task before you switch everything. I track this kind of thing on Goldie Bench.
In ChatGPT Work and Codex, OpenAI says your plan's usage allowance covers substantially more tasks with Terra than with Sol or Astra. If you keep hitting limits, that alone is a reason to make Terra your default for routine work. Want help mapping your workflows to the right model? Book a free AI strategy session.
Which one to use in Codex and Hermes Agent
Codex: OpenAI's model guide lists Terra in Work and Codex in the desktop app, ChatGPT Work on the web and the Codex CLI. Sol is also listed for the Codex IDE extension, and Terra isn't listed there. In the desktop app, open Advanced and choose Terra from the model list. In the CLI, run /model and pick it. Use Terra for quick edits, screenshots and writing tasks, and switch to Sol for real refactors and multi-file builds.
Hermes Agent: the Hermes provider docs support both the OpenAI API (provider openai-api) and a ChatGPT or Codex subscription login (provider openai-codex), but they don't name Terra or Sol yet. They do note that Codex login models get a 272K context window for most slugs, not the 1.05M you get through the direct API. So if you want the full context, use your API key. I run Hermes daily with a profile per model, so Terra gets its own profile for screen work and Sol gets one for code.
Not sure how to split your agents across models? Get a free AI strategy session and we'll map it out. And if you want my exact profiles and prompts, they're inside the AI Profit Boardroom.
FAQ
Is GPT-6.1 Terra better than GPT-6.1 Sol?
For speed, price and many computer-use and writing tasks, yes. Terra costs half as much per token and scores higher than Sol on Agents' Last Exam. For complex coding and long desktop workflows, Sol scores higher. For more on Sol itself, see my full guide to GPT-6.1 Sol.
How much does GPT-6.1 Terra cost?
OpenAI lists $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Sol is $2, $0.10 and $10. Microsoft's Global Standard price in Foundry matches OpenAI's at $1 input and $5 output.
How does Terra compare to GPT-6 Astra?
Astra is OpenAI's top model at $10 input and $50 output per million tokens. OpenAI says Terra performs on par with Astra on Agents' Last Exam at one-tenth the price per token, and finishes tasks over three times faster. Astra still leads on the hardest coding and science work. Here's how to use GPT-6 Astra.
Can I use Terra in ChatGPT right now?
Yes, if you're on Plus, Pro, Enterprise or Edu. OpenAI says it's in ChatGPT Work and Codex, and in the API as gpt-6.1-terra. Pick it from the model list and run one of your daily tasks on it. That's the fastest way to settle GPT 6.1 Terra vs Sol for your own work.
