GPT-6.1 Terra: What's New, Price, and the Best Free Alternative

2026-10-27 · 6 min read

openaigptmodel-releaseai-tools

If your workload is high-volume computer use — reading invoices, filling in forms, turning a technical drawing into a 3D model — GPT-6.1 Terra is the model to test first. OpenAI released it on October 26, 2026 at $1 per million input tokens and $5 per million output tokens, half the price of both GPT-5.6 Terra and GPT-6.1 Sol. OpenAI says it matches GPT-6 Astra on agentic computer tasks at roughly a tenth of Astra's price per token, and it cut its internal factual-error rate by more than half. If your job is the hardest few percent of coding, stay on GPT-6.1 Sol or GPT-6 Astra.

What OpenAI changed

Terra is the efficiency release, not the new flagship. OpenAI describes it as a fast model that combines detailed visual reasoning, precise computer use and clearer, more natural writing, and positions it as a step up from GPT-5.6 Terra rather than a replacement for Astra or Sol.

The three claims that matter:

Agentic computer use. On Agents' Last Exam, a benchmark of professional tasks such as converting architectural drawings into 3D models and logging an animated film shot by shot, Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. OpenAI says it performs on par with GPT-6 Astra at only one-tenth Astra's price per token.
Factuality. On OpenAI's internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra — a drop of nearly 37 percentage points. At maximum effort the error rate fell by 18 points to 4.3%, close to Astra and Sol.
Speed. OpenAI says Terra completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol, and at max effort it is over three times as token-efficient as Claude Opus 5.5 while costing one-fourth as much per token — about 9x lower cost per task.

On coding, the headline is less dramatic. Terra solves 50.3% of Terminal-Bench 4.0 at maximum reasoning effort, more than double GPT-5.6 Terra's published 21.5%, but OpenAI explicitly says Sol and Astra remain the better choice for the most complex coding tasks. On Terminal-Bench-Science 0.1 it surpasses GPT-6.1 Sol at lower cost and comes within five points of Claude Opus 5.5 at less than one-fifth the estimated cost.

One caveat before you rewrite a production config: these are OpenAI's own numbers, published on announcement day. Independent third-party runs take a few days to appear. Treat the direction as reliable and the exact margins as provisional.

The price, in full

Model	Input / 1M	Cached input / 1M	Output / 1M	Context
GPT-6.1 Terra	$1.00	$0.05	$5.00	922,000
GPT-6.1 Sol	$2.00	$0.10	$10.00	—
GPT-6 Astra	$10.00	$1.00	$50.00	—

Terra's API identifier is gpt-6.1-terra, maximum output is 128,000 tokens, and the context window is 922,000 tokens. Three pricing details decide the real bill:

Cache reads cost $0.05 per million, a 95% discount. Replaying a long system prompt or a fixed document on every call is where the saving actually sits, not in the headline price.
Crossing 272,000 input tokens reprices the whole request. Above that line, input costs $2.00 and output $7.50 per million. A 900,000-token request is not nine times a 100,000-token one; it is much more. The cheap window is effectively 272K.
Batch requests are half price at $0.50 input and $2.50 output per million. Anything that does not need a real-time answer should go through the batch endpoint.

Where Terra wins, and where it does not

Use Terra for:

High-volume document and invoice extraction, form filling, screenshot review, and other visual-plus-action loops. This is the workload OpenAI optimized for, and the cost gap against Astra is large enough to change whether a project is viable at all.
Drafting and rewriting at scale. OpenAI positions the writing quality as a step up, and at these prices you can afford to generate and discard more candidates.
Any agent whose bottleneck is latency. Three times faster than Astra and Sol changes what a user will sit through.

Keep Sol or Astra for:

The hardest coding tasks. OpenAI says so directly, and Terminal-Bench 4.0 at 50.3% is good but not frontier-level.
Decisions where "nearly matches Astra" is not good enough. Medical, legal and financial visual review still belong on the model that wins outright, because a human will not re-check every step.
Very long single requests. The 922K window is real, but the 272K repricing cliff makes it expensive to use the whole thing.

Can you use GPT-6.1 Terra for free?

Not on OpenAI's side. Terra ships inside ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and through the paid API. OpenAI did not announce a free ChatGPT tier for it on launch day, so anything advertising "GPT-6.1 Terra free" as an official OpenAI path on October 27 should be checked carefully.

The practical free route is the same as with previous releases: compare the workload on models you can already use without a new subscription, and only pay for Terra if the comparison shows it is the one that clears your bar.

Try it on aiwith.chat

Being straight with you about what is live on our side: GPT-6.1 Terra is not in our model picker yet. We add models after testing them rather than on announcement day, and we will not claim it is available before it is.

What you can use today on aiwith.chat is the comparison that decides the switch: GPT-6.1 Sol, GPT-6 Astra and GPT-6 Luna under one account, alongside the other model families on the site.

Run your real task on Sol and on Astra:

If Sol already clears your quality bar, Terra — cheaper, faster, and benchmarked near Astra — is very likely the better default when it lands.
If only Astra clears it, you have found the jobs worth keeping on a flagship. Terra's "on par with Astra" claim is about computer-use tasks specifically, not a blanket guarantee.

Either way you end up with a short list of jobs that actually need the expensive model, which is the useful output regardless of which release you adopt.

Bottom line

GPT-6.1 Terra is the release that makes high-volume computer use and visual document work cheap: half of Sol's price, about a tenth of Astra's cost per token on agentic tasks, a much lower error rate than GPT-5.6 Terra, and three times the speed. It is not the new coding flagship and not a free model. Test it on the repetitive visual work first, keep the hardest reasoning on Astra or Sol, and do not use the full 922K context unless the request genuinely needs it.