# GPT-6.1 Terra prices computer-use agents at a tenth of Astra, on OpenAI's own scorecard

URL: https://www.thedeepfeed.ai/posts/2026-10-26-gpt-6-1-terra-cheap-tier-for-agents/
Category: Agents
Published: 2026-10-26
Author: the-deep-feed
Tags: openai, gpt-6-1-terra, computer-use, pricing, agents, microsoft-foundry, benchmarks
Kind: news

> OpenAI released GPT-6.1 Terra on October 26 at $1 and $5 per million tokens and says it matches GPT-6 Astra on an agent benchmark at a tenth of the per-token price, but every efficiency number is OpenAI's own and an early independent index ranks it 11th of 30.

## TL;DR

- **GPT-6.1 Terra** lists at **$1** per million input tokens and **$5** per million output, half of GPT-6.1 Sol and a tenth of GPT-6 Astra, and is live in ChatGPT, Codex, the API and Microsoft Foundry from **October 26**.
- OpenAI says Terra matches **Astra** on Agents' Last Exam at a tenth of the per-token price and runs **~9x** cheaper per task than Claude Opus 5.5 at max effort. All of those figures are OpenAI's.
- An early outside index, **Vals AI**, ranks Terra **#11 of 30** at **60.47%**, and OpenAI's own post says Sol and Astra 'remain the better choice' for the hardest coding work.
- The **1,050,000-token** window carries a cliff: past **272K** input tokens the whole request bills at **2x** input and **1.5x** output, and EU data-zone deployments on Foundry cost **20%** more.

GPT-6.1 Terra makes a new low-price tier OpenAI's pitch for computer-use agents: $1 per million input tokens, a tenth of GPT-6 Astra, with OpenAI claiming Astra-level results on an agent benchmark. The claim rests entirely on OpenAI's own measurements, and an early independent index puts Terra in the middle of the field, not at the top.

OpenAI [announced Terra](https://openai.com/index/introducing-gpt-6-1-terra/) on Monday, October 26, as "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing." It went live the same day for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as `gpt-6.1-terra`. Microsoft made it [generally available in Foundry](https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/gpt-6-1-terra-now-generally-available-in-microsoft-foundry-build-agents-that-see/4567769) the same day, under the headline "build agents that see and act."

# A tenth of Astra per token

The list price is $1 per million input tokens, $0.05 for cached input and $5 per million output tokens. OpenAI says that is 50% below both GPT-5.6 Terra and GPT-6.1 Sol. Against the rest of the current lineup the gap is wider. This publication [reported Sol](/posts/2026-10-09-gpt-6-1-sol-prices-speed-above-astra/) at $2 and $10, and [Astra](/posts/2026-09-04-gpt-6-astra-launch/) launched at $10 and $50. Terra therefore costs half of Sol and a tenth of Astra on every token.

The [API model page](https://developers.openai.com/api/docs/models/gpt-6.1-terra) lists a 1,050,000-token context window, 128,000 maximum output tokens and a June 1, 2026 knowledge cutoff. Reasoning effort runs from `none` through `low`, `medium` (the default), `high`, `xhigh` and `max`. Batch and Flex processing cost 50% less than standard; Fast mode costs twice as much.

# The agent claims are self-scored

OpenAI's headline argument is about agents, and it is made on numbers OpenAI produced. On Agents' Last Exam, the post says, "At medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. It even performs on par with GPT-6 Astra at only one-tenth its price per token." At max effort, it adds, Terra "is over three times as token efficient as Claude Opus 5.5, and one-fourth its price per token, resulting in a ~9x lower cost per task." It also says Terra finishes tasks "over three times faster" than Astra and Sol.

The reliability claim is the one agent builders will test first. On an internal set of prompts designed to induce errors, OpenAI says 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 max effort fell 18 percentage points to 4.3%. The benchmark is internal and unpublished, so outsiders cannot rerun it. For MMMU-Pro, OSWorld 2.1 and Terminal-Bench-Science 0.1 the post describes gains in words and gives no numbers.

The cost-per-task framing deserves the most care. A "~9x lower cost per task" against Opus 5.5 multiplies two OpenAI-measured quantities, token efficiency and price, on a benchmark OpenAI chose. Per-task cost is the right unit for agents, because an agent that needs three times the tokens erases a cheap rate. The number is only as good as the token counts behind it, and those have not been reproduced.

# What an outside index found

An early third-party read came from [Vals AI](https://www.vals.ai/models/openai_gpt-6.1-terra), which ran Terra "via OpenAI's API with reasoning effort max." Its summary: "It ranks #11 of 30 models on the Vals Index with 60.47%." Terra's best placement there is "#20 of 122 on Vibe Code Bench v1.1," and Vals lists its cost per test at $2.925.

A mid-table rank does not contradict OpenAI's claim, which was about price for a given level of agent performance, not about leading the field. It does put a boundary on it. OpenAI draws the same boundary on coding. Terra solves 50.3% of Terminal-Bench 4.0 tasks at maximum effort, more than double GPT-5.6 Terra's 21.5%, but the post concedes: "For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice."

The [Hacker News thread](https://news.ycombinator.com/item?id=50248124) on launch day reached 636 points and 427 comments, and it split along that line. One commenter read the charts at face value:

> Wow. If those charts are true, this effectively matches GPT-6-Astra at a fraction of the cost, and is neck-and-neck with 6.1-Sol.
>
> — A_D_E_P_T, Hacker News, Oct 26, 2026

Another pointed at the coding table:

> That terminal bench 4.0 result is... bad? Terra XHigh more expensive than Sol Low and still does worse.
>
> — Starlevel004, Hacker News, Oct 26, 2026

# The fine print on the million-token window

Two pricing details change the arithmetic for agent workloads, which tend to carry long histories.

The first is a context cliff. OpenAI's model page says "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." The surcharge applies to the whole call, not the tokens past the line. A 500,000-token prompt with a 10,000-token answer costs about $1.08 at the long-context rate, against $0.55 if the same tokens had been billed at the standard rate. Agents that let their transcript grow past 272K pay double on every subsequent turn. Microsoft's Foundry table shows the same structure, with long-context Global Standard at $2.00 input and $7.50 output.

The second is residency. Foundry's Global Standard price matches OpenAI's $1 and $5. Data Zone deployments, which keep processing inside a region, cost $1.10 and $5.50 in the US and $1.20 and $6.00 in the EU and APAC. A European enterprise that must keep agent traffic in-region pays 20% above the headline rate.

# The safety rating travels with the price cut

Terra is cheap but not lightly gated. OpenAI says it "meets the Critical threshold in cybersecurity under our Preparedness Framework, and maintains the protections developed for GPT-6 Astra." In September, [Astra was the first model](/posts/2026-09-04-gpt-6-astra-launch/) OpenAI rated Critical in that category. Less than eight weeks later, a model in that category is selling at a tenth of Astra's price, which makes the protections, not the price, the main thing that limits what cheap agents can do with it.

The open question is whether OpenAI's agent numbers hold up outside its own harness. A tenth of Astra's price for Astra's agent results would make Terra the default for high-volume computer use. At Vals AI's #11 of 30, with OpenAI itself recommending Sol and Astra for hard coding, Terra currently prices like a default and benchmarks like a mid-tier model, and teams should rerun their own task suites before switching on that claim.

## Sources

- [OpenAI — Introducing GPT-6.1 Terra (Oct 26, 2026)](https://openai.com/index/introducing-gpt-6-1-terra/)
- [OpenAI API docs — GPT-6.1 Terra model page](https://developers.openai.com/api/docs/models/gpt-6.1-terra)
- [Microsoft Foundry blog — GPT-6.1 Terra now generally available in Microsoft Foundry (Oct 26, 2026)](https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/gpt-6-1-terra-now-generally-available-in-microsoft-foundry-build-agents-that-see/4567769)
- [BleepingComputer — OpenAI launches GPT-6.1 Terra with 3x faster AI agents, 50% lower pricing (Oct 26, 2026)](https://www.bleepingcomputer.com/news/artificial-intelligence/openai-launches-gpt-61-terra-with-3x-faster-ai-agents-50-percent-lower-pricing/)
- [Vals AI — GPT-6.1 Terra benchmarks, cost and capabilities](https://www.vals.ai/models/openai_gpt-6.1-terra)
- [Hacker News — GPT-6.1 Terra discussion (Oct 26, 2026)](https://news.ycombinator.com/item?id=50248124)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-10-26-gpt-6-1-terra-cheap-tier-for-agents/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt