GPT-6.1 Terra vs Luna, Sol, Astra: which tier to buy

OpenAI released GPT-6.1 Terra on Oct 26, 2026. It’s a mid-sized model that the company says is faster, more accurate and half the API price of the GPT-5.6 Terra it replaces. If you buy API capacity, the useful question is where Terra sits among OpenAI’s four tiers, and what you give up by not paying for the ones above it.

One caveat covers everything below: the speed, cost and error-rate figures are OpenAI’s own, from its announcement and internal benchmarks. No independent testing has been reported yet.

How do Luna, Terra, Sol and Astra line up?

OpenAI’s lineup has four classes. Luna is the cheapest and least capable, Astra is the priciest and most capable, and Terra is the second rung, with Sol above it. The current models named in coverage are GPT-6 Luna, GPT-6.1 Terra, GPT-6.1 Sol and GPT-6 Astra.

• Luna: the cheapest, least capable class. Android Authority says Terra has significantly lower error rates than GPT-6 Luna. Luna’s price isn’t in the coverage.
• Terra: the new mid-tier model, aimed at visual reasoning, computer use and writing.
• Sol: one step up. Android Authority says GPT-6.1 Sol arrived about a month ago.
• Astra: the top class. Android Authority says unnamed concerns have disrupted plans to launch a 6.1 version. It gives no detail.

Outlets disagree on what shipped last month. 9to5Mac says GPT-6 Sol and Luna were released and Terra was left out. Android Authority says GPT-6.1 Sol. Check exact model names in OpenAI’s docs before you hard-code anything.

What does GPT-6.1 Terra cost?

BleepingComputer lists $1 for every million input tokens and $5 for every million output tokens. That’s the only source with those figures. OpenAI says Terra’s API price is half that of GPT-5.6 Terra, and that Terra costs about one-tenth of GPT-6 Astra per token.

The 50% figure is where the coverage gets muddy. 9to5Mac and Android Authority measure the cut against GPT-5.6 Terra. BleepingComputer says the cut is against both GPT-5.6 Terra and GPT-6.1 Sol. Android Authority also says Terra is roughly half Sol’s price, then elsewhere calls it only slightly cheaper than Sol. Those can’t all be right.

Until OpenAI’s pricing page settles it, I wouldn’t build a budget on a specific Terra-versus-Sol gap. Prices for Luna, Sol and Astra aren’t in the coverage at all.

Where Terra is probably enough

For agent-style work, Terra looks like the default to test first. OpenAI says it combines detailed visual reasoning, precise computer use and clearer writing. The examples it gives are drafting proposals, interpreting technical drawings, filling out forms and editing videos.

The numbers OpenAI offers, all self-reported:

• On Agents’ Last Exam at medium reasoning effort, Terra beat GPT-5.6 Terra by 15 percentage points, at roughly a third of the cost.
• On OpenAI’s own test built from prompts meant to provoke mistakes, Terra made errors on 16.7% of prompts at low reasoning effort, compared with 53.6% for GPT-5.6 Terra. At maximum effort the figure is 4.3%, down 18 points.
• At maximum effort, OpenAI says Terra’s error rate nearly matches GPT-6 Astra and GPT-6.1 Sol.
• OpenAI says Terra finishes tasks over three times faster than Astra and Sol. BleepingComputer states it that broadly. Android Authority limits it to computer-use tasks.

That last point matters if your workload is mostly text generation rather than agents. Test the speed on your own jobs.

Where Sol and Astra still earn their price

Hard coding is the clearest case. For the hardest coding jobs, OpenAI continues to point customers to GPT-6.1 Sol or GPT-6 Astra, according to BleepingComputer. That’s the company steering its own customers upward.

Terra’s Terminal-Bench 4.0 result at maximum effort is 50.3%, against 21.5% for GPT-5.6 Terra, more than double. But that compares Terra to its predecessor. The coverage doesn’t give Sol or Astra scores on the same test, so you can’t tell how big the remaining gap is.

The Register took a skeptical line, calling the release no bombshell while granting that Terra competes with costlier models on certain tasks for roughly a tenth of what Astra’s API charges. The words doing the work there are “some tasks.” OpenAI hasn’t said which ones.

Luna is the harder call. Terra beats it on error rate, per Android Authority. If Luna’s price advantage is large for your volume, measure its mistakes on your data before you rule it in or out.

What it means for your API budget

My suggestion: make Terra your baseline and make Sol or Astra prove they’re worth the premium. Build a test set from your real prompts, including the ugly edge cases, and run it at low and maximum reasoning effort. OpenAI’s own numbers show a huge swing between those settings (16.7% versus 4.3% error for Terra), so effort level is a cost lever you’ll want to price in.

If you haven’t built an evaluation habit, a well-regarded book on production ML, such as Chip Huyen’s Designing Machine Learning Systems, is a reasonable place to start. Vendor benchmarks are only a starting point.

Access is straightforward. Terra is available now through the API as gpt-6.1-terra, and Plus, Pro, Enterprise and Edu users can reach it inside ChatGPT Work and Codex. Free ChatGPT access isn’t mentioned in the coverage.

If you’re on GPT-5.6 Terra, there’s no retirement date reported yet. Still, a model that OpenAI says is half the price with a lower error rate is worth a trial run this week.

If you’re comparing beyond OpenAI, our scorecard of Meta Muse Wonder against Opus, Astra and Gemini 4 puts several rival models side by side. For context on what AI vendors spend to run this business, see our explainer on the loss disclosed in Anthropic’s IPO filing.

What nobody has verified yet

Independent testing hasn’t been reported, so the error-rate, speed and Terminal-Bench figures are all OpenAI’s word. How its mistake-provoking prompt test is built is also unexplained, and it may not reflect real-world hallucination rates.

Other gaps:

• Which tasks the “on par with Astra at one-tenth the price” claim covers.
• Whether the 50% cut is measured against GPT-5.6 Terra only, or against Sol too.
• The prices of Luna, Sol and Astra.
• What ChatGPT Work includes and how it differs from regular ChatGPT.
• What concerns held up a GPT-6.1 version of Astra.
• Whether and when GPT-5.6 Terra leaves the API.

The $1 and $5 token prices come from a single outlet, so confirm them on OpenAI’s pricing page before you commit.

Frequently asked questions

How much does GPT-6.1 Terra cost in the API?

BleepingComputer reports $1 for every million input tokens and $5 for every million output tokens. It's the only source with those numbers, so confirm them on OpenAI's pricing page. OpenAI says the price is 50% lower than GPT-5.6 Terra.

Is GPT-6.1 Terra better than GPT-6.1 Sol?

Not across the board, according to OpenAI. It says Terra's maximum-effort error rate nearly matches Sol's and that it's over three times faster. But OpenAI still recommends Sol or GPT-6 Astra for the most complex coding tasks.

What is the API model name for GPT-6.1 Terra?

The API name is gpt-6.1-terra. Plus, Pro, Enterprise and Edu users can also use it in ChatGPT Work and Codex.

Can free ChatGPT users use GPT-6.1 Terra?

Not yet reported. Coverage lists only paid tiers: Plus, Pro, Enterprise and Edu.

Sources

This article was compiled from reporting by the following outlets. Links go to the original reports.

1. 9to5Mac: OpenAI upgrades ChatGPT and Codex with new GPT-6.1 Terra model
2. Neowin: OpenAI releases GPT-6.1 Terra with impressive performance at massively reduced costs
3. BleepingComputer: OpenAI launches GPT-6.1 Terra with 3x faster AI agents, 50% lower pricing
4. The Register: OpenAI rolls out budget-priced mid-tier GPT-6.1 Terra
5. Android Authority: GPT-6.1 Terra makes a strong case for being OpenAI’s best bang for your AI buck