GPT-6.1 Terra: Release Date, Price, and How It Compares to Sol

OpenAI released GPT-6.1 Terra on October 26, 2026, replacing GPT-5.6 Terra as the mid-tier model in ChatGPT Work, Codex, and the API. It is pitched as a fast, efficient model built for visual reasoning and computer use, with standard API pricing 50% below both GPT-5.6 Terra and GPT-6.1 Sol — at $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens.

What is GPT-6.1 Terra?

In its launch post, OpenAI describes GPT-6.1 Terra as "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing." The company calls it "a significant step up from GPT-5.6 Terra," built for the communication and computer-based tasks that fill a typical workday — drafting proposals, interpreting technical drawings, filling out forms, and editing videos.

Terra sits in the middle of OpenAI's current model lineup. From lowest to highest capability, the family now runs GPT-6 Luna, GPT-6.1 Terra, GPT-6.1 Sol, and GPT-6 Astra, OpenAI's most capable frontier model. Terra's job in that lineup hasn't changed with this release — it's still the tier aimed at high-volume, everyday work rather than the hardest reasoning or coding problems — but OpenAI says it has moved the "cost-intelligence frontier" for that tier considerably further than the GPT-5.6 version did.

Release date and where to get it

GPT-6.1 Terra went live on October 26, 2026. OpenAI's own ChatGPT release notes list it as rolling out the same day to ChatGPT Work and Codex, covering Plus, Pro, Business, Enterprise, and Edu plans. Developers can call it through the API immediately under the model ID gpt-6.1-terra, via both the Responses API and Chat Completions. One access caveat worth knowing: OpenAI's platform changelog and release notes flag that Terra does not currently support UAE data-residency inference, a restriction tied to where the model runs rather than to the user's own location.

Inside ChatGPT Work and Codex, OpenAI says a plan's usage allowance "goes further" with GPT-6.1 Terra — the same monthly allowance now covers more Terra tasks than it did with GPT-5.6 Terra, and substantially more than the same allowance buys with GPT-6.1 Sol or GPT-6 Astra. That's a direct result of Terra's lower per-token cost: cheaper requests mean more of them fit inside a fixed usage budget.

Pricing: how much cheaper is it?

The headline number is a 50% price cut. OpenAI's API pricing page and the model's own documentation page list standard rates of $1.00 per million input tokens, $0.05 per million cached input tokens, and $5.00 per million output tokens — half of what GPT-5.6 Terra and GPT-6.1 Sol charge for the same categories. A few more pricing details worth noting for anyone building on the API:

• Cache writes cost $1.25 per million tokens with 30-minute retention.
• Prompts longer than 272,000 input tokens are billed at 2x the input and cache rates and 1.5x the output rate.
• Batch and Flex processing modes cut the standard price by a further 50%.
• Regional data processing adds a 10% premium; "fast" mode costs 2x the standard rate.

OpenAI frames the cut as making Terra "practical for developers to automate high-volume work," pointing specifically to tasks like reading incoming invoices and entering their details into accounting software — the kind of repetitive, high-request-volume job where a 50% per-token discount compounds quickly. For ChatGPT and Codex subscribers who aren't paying per token, the practical effect of the price cut shows up as that larger effective usage allowance rather than a bill.

The price cut also changes the economics of agentic workloads specifically. OpenAI's own cost-per-task framing on the Agents' Last Exam benchmark is a useful way to see this: Terra at max reasoning effort is roughly a quarter of Claude Opus 5.5's price per token and more than three times as token-efficient on that benchmark, which OpenAI translates into about a 9x lower cost to complete the same task. For a workload run thousands of times a day — invoice processing is OpenAI's own example — that kind of multiplier matters more than the headline per-token price.

What's actually new: the benchmark numbers

OpenAI backs the "significant step up" claim with results across four benchmarks, each measured against GPT-5.6 Terra specifically:

• Agents' Last Exam (professional computer-use tasks, like turning architectural drawings into a 3D model): at medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost, and performs on par with GPT-6 Astra at roughly a tenth of Astra's price per token. At max effort, OpenAI says Terra is more than three times as token-efficient as Claude Opus 5.5 and about a quarter of its price per token — a roughly 9x lower cost per completed task. Terra also finishes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.
• MMMU-Pro (visual reasoning): Terra beats GPT-5.6 Terra at every reasoning setting and comes close to GPT-6 Astra and GPT-6.1 Sol's scores at a much lower cost.
• OSWorld 2.1 (multistep desktop-app tasks): at medium effort, Terra surpasses GPT-5.6 Terra's best score at a fraction of the cost per task.
• Terminal-Bench 4.0 (command-line coding tasks): at max effort, Terra solves 50.3% of tasks versus GPT-5.6 Terra's published 21.5% — more than double. OpenAI is upfront that GPT-6.1 Sol and GPT-6 Astra "remain the better choice" for the hardest coding work.

OpenAI also reports a sharp drop in factual errors on its internal set 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, and at maximum effort it fell a further 18 points to 4.3%, which OpenAI says nearly matches the error rates of GPT-6 Astra and GPT-6.1 Sol. On Terminal-Bench-Science 0.1, a test of using code for research tasks, Terra at max reasoning surpasses GPT-6.1 Sol at a lower cost and comes within five points of Claude Opus 5.5 while costing under a fifth as much, by OpenAI's estimate. For more on how Sol itself performs against that same Anthropic model, see our benchmark comparison of Claude Opus 5.5 and GPT-6.1 Sol.

GPT-6.1 Terra vs. Sol, Astra, and Luna

With Terra refreshed, all four tiers of OpenAI's current lineup are now easier to compare directly. Here's how OpenAI positions each one, based on its own pricing and model pages:

The practical read: Terra is no longer just "the cheap option that falls behind." On visual reasoning (MMMU-Pro) and on the Agents' Last Exam benchmark at medium effort, OpenAI's own numbers show it closing in on Astra and Sol while costing a fraction as much and running several times faster. Sol and Astra still win outright on the hardest coding and reasoning tasks — Terminal-Bench 4.0 makes that explicit — so the decision mostly comes down to whether your work looks like drafting, form-filling, invoice processing, and screen-based tasks (Terra's strength) or deep multi-step coding and research (where Sol and Astra remain ahead). We covered how Sol and Astra differ from each other, including pricing and access, in our earlier breakdown of GPT-6.1 Sol vs. Astra.

Visual reasoning and computer use, in practice

The two capabilities OpenAI leans on hardest for Terra are visual reasoning and computer use. In practice, that means Terra is designed to look at an image — a technical drawing, a scanned form, a video frame — and act on what it sees, rather than just describe it. OpenAI's examples include turning architectural drawings into 3D models, reviewing an animated film shot by shot to build a time-coded log, filling out forms, and editing videos. On the writing side, OpenAI also highlights "clearer, more natural writing" as an improvement over GPT-5.6 Terra, relevant for the proposal-drafting and documentation work the model is aimed at.

Technically, Terra ships with a 1,050,000-token context window and a 128,000-token maximum output, accepts text and image input (not audio or video input), and has a knowledge cutoff of June 1, 2026, according to its platform documentation. Reasoning effort is adjustable across six levels — none, low, medium (the default), high, xhigh, and max — which is how OpenAI gets such different cost and accuracy numbers out of the same model in its benchmark tables. Tool calling requires the Responses API rather than Chat Completions.

That effort dial is really the headline feature for anyone deciding how to deploy Terra. At low or medium effort, it's the cheapest, fastest option in the lineup for form-filling, document review, and drafting; dialed up to max effort, OpenAI's own Agents' Last Exam and Terminal-Bench-Science numbers show it reaching into territory that used to require Sol or Astra, just at a noticeably higher per-task cost than Terra's low-effort mode. In other words, Terra isn't really one model with one price — it's a model whose cost and capability both scale with the effort setting you choose for a given task.

Safety and alignment

OpenAI says GPT-6.1 Terra meets the "Critical" threshold for cybersecurity under its Preparedness Framework and retains the safety protections built for GPT-6 Astra, while also carrying over the alignment improvements made across the broader GPT-6 series. The company's system card addendum for GPT-6.1 Terra covers these evaluations in more detail, including testing for prompt-injection resistance and for the model's tendency to stay in scope and disclose its own limitations — both areas where OpenAI says Terra improves on GPT-5.6 Terra.

Which tier should you actually use?

For most people inside ChatGPT, the honest answer is: let the model picker handle it, since usage allowances now stretch further with Terra specifically. But if you're choosing deliberately, or building against the API where the tier is a real cost decision, the OpenAI numbers above point to a fairly clean split:

• Pick GPT-6.1 Terra if the work is visual-reasoning-heavy, computer-use-heavy, or high-volume and repetitive — reading documents, processing invoices, filling forms, drafting proposals, light coding and desktop-agent tasks. It's the cheapest tier that OpenAI's own benchmarks show closing in on Sol and Astra for these jobs.
• Pick GPT-6.1 Sol or GPT-6 Astra for the hardest coding problems, multi-step research, or reasoning tasks where Terminal-Bench 4.0 and Terminal-Bench-Science 0.1 still show a real capability gap. If you're an Anthropic customer weighing Claude Opus 5.5 against this tier of OpenAI model, our Opus 5.5 vs. GPT-6.1 Sol comparison covers that matchup directly.
• Pick GPT-6 Luna for quick, low-stakes chat where neither visual reasoning nor computer use is in play and you want the lowest possible cost or want to stay on a free plan.

This release also follows directly from the model-family changes OpenAI introduced around DevDay 2026, where GPT-6.1 Sol and the Ultrafast mode first appeared — Terra's refresh is the next step in filling out that same four-tier lineup rather than a standalone launch.

What's next

With Terra refreshed, GPT-6.1 Sol is now the only tier still running its original GPT-6.1-era benchmarks and pricing without a direct successor announced; based on OpenAI's release cadence this year, a comparable Sol refresh or a further Astra update would be the next logical move, though OpenAI hasn't announced a date for either. Worth watching too: whether the UAE residency gap for Terra gets resolved, and whether the usage-allowance math changes again as more ChatGPT Work and Codex users shift their default model to Terra. We'll update this page if OpenAI publishes pricing, benchmark, or availability changes for any tier in the lineup.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage