GPT-6.1 Terra Explained: Price, Speed, and How to Use It

GPT-6.1 Terra Explained: Price, Speed, and How to Use It

OpenAI quietly dropped a model on October 26, 2026 that could change what it costs to put an AI agent to work. It's called GPT-6.1 Terra, and if you run anything that makes an AI click buttons, fill in forms, read screenshots, or grind through repetitive computer tasks, this is the launch to pay attention to. Not because it's the smartest model OpenAI has ever shipped — it isn't — but because it's fast, cheap, and built for exactly the boring work that eats up real workdays.

In this guide, I'll explain what GPT-6.1 Terra actually is, what it costs, how it stacks up against GPT-6.1 Sol, GPT-6 Astra, and Claude Opus 5.5, and how you can try it today in ChatGPT Work, Codex, or the API. No hype, just the numbers OpenAI published — and the honest caveats behind them.

Quick Answer: What Is GPT-6.1 Terra?

GPT-6.1 Terra is OpenAI's new mid-tier model for visual reasoning and computer use, launched October 26, 2026. It replaces GPT-5.6 Terra, costs $1 per million input tokens and $5 per million output tokens through the API (about half the price of GPT-6.1 Sol), and is available to paid users in ChatGPT Work and Codex, plus developers via the API as gpt-6.1-terra. OpenAI says it completes computer tasks more than three times faster than its bigger siblings — though, as we'll see, those speed and benchmark claims come from OpenAI's own tests.

Table of Contents

What GPT-6.1 Terra Is Built For

OpenAI describes Terra as "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing." In plain English: it looks at a screen, understands what's on it, and then does something about it — clicking, typing, moving between apps, and filling things in.

The examples OpenAI chose are deliberately unglamorous, and that's the tell. We're talking about logging incoming invoices into accounting software, filling out forms, drafting proposals, reading technical drawings, and editing video. This isn't a model pitched at research breakthroughs. It's pitched at the repetitive work that eats an afternoon.

That focus makes sense when you look at where OpenAI's lineup sits. From cheapest to most capable, the family runs Luna, Terra, Sol, and Astra. Terra is the middle slot: strong enough for real professional work, cheap enough to run at volume. If you've followed OpenAI's recent releases, you'll remember that GPT-6's intelligent UI pushed ChatGPT toward doing work inside apps rather than just chatting about it — Terra is the engine-room version of that same idea, aimed at developers and teams automating screens, documents, and forms.

Computer use, in practice

Computer use means the model works from what it sees on screen rather than from a clean API feed. It reads a screenshot, finds the right field, types the value, checks the result, and moves on. OpenAI's benchmark for this, Agents' Last Exam, tests exactly that kind of professional task — turning architectural drawings into a 3D model, or building a time-coded log from an animated film.

Terra also handles visual reasoning well beyond screenshots: matching product photos to catalog entries, interpreting dense diagrams, and reading technical drawings are all listed use cases. If your work involves images plus actions — not just images plus answers — Terra is built for that combination.

Key specs at a glance

  • Context window: 1,050,000 tokens, with up to 128,000 output tokens
  • Input: text and images (no native audio or video input)
  • Knowledge cutoff: June 1, 2026
  • Reasoning effort: from none to max, with medium as the default
  • Tools: computer use, hosted shell, code interpreter, web search, file search, and MCP (tool calling requires the Responses API)
  • Availability: ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu plans; API as gpt-6.1-terra

GPT-6.1 Terra Pricing Explained

Here's the number that made everyone stop scrolling: $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens at standard rates. OpenAI says that's roughly 50% lower than GPT-5.6 Terra and GPT-6.1 Sol — and on output tokens, where agent bills usually pile up, the cut is even deeper than 50% against the old Terra's $12.

But token price isn't the same as task price, and OpenAI's own fine print matters:

  • Long prompts cost more: prompts above 272,000 input tokens are charged at double input and cache rates, and 1.5x the output rate for the full request.
  • Cache writes aren't free: 30-minute cache writes cost $1.25 per million tokens.
  • Batch and Flex processing cost 50% less than standard rates, while Fast mode costs twice as much.
  • Regional processing adds 10% where available.

Still, the per-task math is striking. On OpenAI's Agents' Last Exam chart, a medium-effort Terra task costs an estimated $0.29, versus $1.31 for GPT-6.1 Sol at the same effort and $4.82 for GPT-6 Astra. At maximum effort, Terra costs about $1.01 per task against $8.82 for Claude Opus 5.5 — roughly a ninth of the price for a score only about three points lower. That's the whole pitch in one line: near-flagship results on computer work, at a fraction of the bill.

One honest caveat: cheaper tokens don't automatically mean cheaper tasks. On one desktop benchmark (OSWorld 2.1), Terra at maximum effort cost $1.19 per task versus $1.15 for Sol — because Terra used more tokens and steps on that particular run. If you're running agents at scale, measure cost per completed task, not just cost per token.

Benchmarks: How GPT-6.1 Terra Compares

Every number in this section comes from OpenAI's own launch materials. No independent lab has published full Terra results yet, so treat these as OpenAI's claims, not settled fact. That said, the pattern across benchmarks is consistent: a big jump over the old Terra, and close to the bigger models on visual and computer-use work.

Benchmark (max effort)GPT-6.1 TerraGPT-6.1 SolGPT-6 AstraClaude Opus 5.5
Agents' Last Exam (computer work)60.0%58.6%59.3%63.2%
OSWorld 2.1 (desktop apps)74.2%78.0%80.9%83.9% (Sonnet 5.5)
MMMU-Pro Vision (image reasoning)85.5%86.3%86.7%—
Terminal-Bench 4.0 (terminal coding)50.3%55.9%58.2%70.6% (Sonnet 5.5)

A few takeaways worth noting:

  • Versus the old Terra, it's a leap. Agents' Last Exam jumped from 42.6% to 57.6% at medium effort (15 points), and Terminal-Bench 4.0 more than doubled, from 21.5% to 50.3% at max effort.
  • On computer work, it beats Sol and Astra. At max effort on Agents' Last Exam, Terra's 60.0% edges out Sol (58.6%) and Astra (59.3%) — while costing far less per task.
  • On hard coding, the bigger models still win. OpenAI itself says Sol and Astra remain the better choice for the most complex coding tasks, and the Terminal-Bench gap backs that up.
  • Factual errors dropped sharply. On OpenAI's internal error-prone prompt set, the error rate fell from 53.6% (old Terra, low effort) to 16.7%, and to 4.3% at max effort — nearly matching Astra and Sol.
  • Speed is a real claim, with a caveat. OpenAI says Terra finishes tasks more than three times faster than Astra and Sol, based on its own timing runs. Independent speed tests haven't landed yet.

For context, this launch sits in a busy month for agentic AI. Anthropic's Claude Duet agent and other agent-focused releases show the whole industry racing toward models that don't just answer — they act. Terra is OpenAI's answer for the high-volume, price-sensitive end of that race. If you want the broader landscape, our roundup of the best AI tools for YouTube creators shows how the same agent trend is reshaping creative work too — and if you create videos with AI, you can grab a ready-made thumbnail in seconds with our free YouTube Thumbnail Downloader.

How to Use GPT-6.1 Terra Today

Option 1: In ChatGPT Work

  1. Open ChatGPT Work (this is where Terra lives — it isn't in regular Chat).
  2. Make sure you're on a Plus, Pro, Business, Enterprise, or Edu plan.
  3. Pick GPT-6.1 Terra from the model list.
  4. Give it a concrete computer task: "Read this invoice PDF and enter the totals into the attached spreadsheet," not "help me with accounting."
  5. Check its work before it touches anything important — computer-use agents can click the wrong button, and Terra is no exception.

Option 2: In Codex

  1. Update the Codex CLI to version 0.159.0 or later.
  2. Run codex -m gpt-6.1-terra to start a session with Terra.
  3. Point it at a well-scoped task — terminal work, form filling, or multi-step desktop jobs — and let medium reasoning effort (the default) do the heavy lifting.
  4. Step up to max effort only when accuracy matters more than speed or cost.

Option 3: Via the API

  1. Use the model ID gpt-6.1-terra in the Responses API (tool calling requires it).
  2. Start with medium reasoning effort and cached prompts where possible — caching at $0.05 per million tokens is where the savings stack up.
  3. Keep prompts under 272,000 input tokens when you can, to avoid the higher long-prompt rates.
  4. Track cost per completed task, not just tokens, before you scale a workflow to thousands of runs.

A practical tip from early testers: Terra shines on focused, well-scoped tasks with clear success criteria. Vague, open-ended jobs ("organize my computer") are where agents wander, burn tokens, and make mistakes. The narrower the job, the better Terra looks.

Who Should (and Shouldn't) Use GPT-6.1 Terra

Great fit for: teams automating invoices, forms, and data entry; developers building screen-based agents; creators and small businesses that process lots of images and documents; anyone currently paying Sol or Astra prices for routine computer work.

Think twice if: you need the hardest coding or research reasoning (Sol and Astra still lead there); you need native audio or video input (Terra is text and image only); you're on a Free plan (OpenAI's announcement names paid plans only); or your workflow needs guaranteed accuracy on high-stakes actions without human review — no agent model is there yet.

There's also a safety note worth knowing: OpenAI classifies Terra at its Critical cybersecurity capability threshold under its Preparedness Framework, with the same protections developed for GPT-6 Astra. That's OpenAI's own classification, and it's a reminder that a model this good at using computers needs guardrails, human approval steps, and limited permissions in any production setup.

Pros and Cons

Pros

  • Half the API price of GPT-6.1 Sol, with better computer-use scores on OpenAI's headline benchmark
  • More than 3x faster task completion than Astra and Sol, according to OpenAI's timing runs
  • Strong visual reasoning (85.5% on MMMU-Pro) close to the flagship models
  • Huge 1.05M-token context window for long documents and multi-step jobs
  • Available right now in ChatGPT Work, Codex, and the API — not a waitlist announcement

Cons

  • All headline benchmarks are OpenAI's own — independent verification is still pending
  • Still trails Sol, Astra, and Claude on the hardest coding tasks
  • Text and image input only — no native audio or video
  • Long prompts, cache writes, and Fast mode carry higher rates that can surprise you
  • Not available on Free plans, and rollout may vary by plan, region, and workspace

Verdict: Should You Switch to GPT-6.1 Terra?

If you pay for AI agents that do computer work, yes — try Terra on your most routine workflow this week. The combination of $1/$5 pricing, near-flagship computer-use scores, and 3x faster completion is exactly the trade most teams have been waiting for: not a smarter model, but a cheaper, faster one that's good enough for the work that actually fills the day.

Keep Sol or Astra for the hardest coding and reasoning. But for invoices, forms, screenshots, catalogs, and drawings — the unglamorous 80% of agent work — Terra just reset what that work should cost. That's why this launch matters more than its benchmark table suggests.

For more on where AI models are heading, check out our explainers on DeepSeek V4.1 Flash, MiniMax M3.1 Flash, and Kimi K3.1 Pro — and keep an eye on The AI Tools homepage and our AI News section for the next launch, because this pace isn't slowing down.

Frequently Asked Questions

What is GPT-6.1 Terra?

GPT-6.1 Terra is OpenAI's mid-tier AI model launched on October 26, 2026, built for visual reasoning and computer use — reading screens, filling forms, and automating everyday office tasks at a lower price than larger models.

How much does GPT-6.1 Terra cost?

Through the API, GPT-6.1 Terra costs $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens at standard rates — about half the price of GPT-6.1 Sol.

How do I access GPT-6.1 Terra?

GPT-6.1 Terra is available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, and through the API as gpt-6.1-terra. It is not in regular Chat, and Free plans are not listed.

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

On OpenAI's computer-use benchmark (Agents' Last Exam), Terra beats Sol at every effort level and costs far less per task. But Sol remains better for the most complex coding tasks, so the best choice depends on your workload.

Is GPT-6.1 Terra free?

No. OpenAI's announcement names paid plans only (Plus, Pro, Business, Enterprise, Edu), and API use is billed per token. There is no free tier listed for GPT-6.1 Terra.

What is GPT-6.1 Terra best used for?

Terra is best for high-volume computer work: entering invoices, filling forms, drafting proposals, reading technical drawings, matching product photos, and editing video — tasks where speed and low cost per task matter most.

Last updated: October 27, 2026. Sources: OpenAI's GPT-6.1 Terra launch announcement and developer documentation, as reported by BleepingComputer, MarkTechPost, and TechStrong AI. Benchmark figures are OpenAI's own measurements and have not yet been independently verified. This article is not sponsored by OpenAI or any company mentioned.



Comments

Post a Comment