In short
- $1.00 input, $0.05 cached input, $5.00 output per 1M tokens (official model page). Cached input is 5% of the input rate.
- The predecessor is GPT-5.6 Terra, not GPT-6.1 Sol. OpenAI's verbatim: "A significant step up from GPT-5.6 Terra". Sol is a separate, higher grade in the same GPT-6.1 generation.
- 1,050,000-token context and 128,000 max output tokens. Knowledge cutoff 2026-06-01 (official model page).
- Above 272,000 input tokens the whole request is repriced at 2x input and 1.5x output. Streaming one huge document is expensive.
- Reasoning effort: none / low / medium (default) / high / xhigh / max. Input: text and image. Output: text.
- Available in ChatGPT Work and Codex (Plus, Pro, Enterprise, Edu) and in the API as
gpt-6.1-terra. - The launch post gives no explicit date ("Starting today"); third-party coverage agrees on October 26, 2026, and we label it as such.
How Terra relates to GPT-6.1 Sol (official wording only)
| Question | Official wording | Our reading |
|---|---|---|
| Predecessor | "A significant step up from GPT-5.6 Terra" | Terra succeeds GPT-5.6 Terra |
| Versus Sol | "Its standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol" | A separate, lower-cost grade in the same GPT-6.1 generation; not a Sol successor |
| When to use which | "For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice." | Sol / Astra for the hardest coding work |
Source: OpenAI — Introducing GPT-6.1 Terra (retrieved 2026-10-29)
Pricing (USD per 1M tokens, official model page)
Standard is quoted as printed. Fast, Batch/Flex prices are our arithmetic from the official multipliers (2x, 50% off).
| Item | Standard | Fast (2x) | Batch / Flex (50% off) |
|---|---|---|---|
| Input | $1.00 | $2.00 | $0.50 |
| Cached input | $0.05 | $0.10 | $0.025 |
| Cache write (1.25x) | $1.25 | $2.50 | $0.625 |
| Output | $5.00 | $10.00 | $2.50 |
Note: cache writes cost $1.25 per 1M tokens for 30-minute retention and $2.50 for 12-hour retention. Batch and Flex are 50% lower than Standard; Fast mode is 2x; regional processing adds 10%.
Source: OpenAI — GPT-6.1 Terra (retrieved 2026-10-29)
The 272K wall: one giant request costs double
Official wording: "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." A 1.05M window does not mean a 1.05M request is cheap.
| Case | Math | Cost |
|---|---|---|
| 300,000 input + 5,000 output in one request | $2/M x 0.3 + $7.5/M x 0.005 (2x input, 1.5x output because the request exceeds 272K) | ~$0.64 |
| Same content split into three 100,000-token calls | $1/M x 0.3 + $5/M x 0.005 (under 272K, so no multiplier; 5,000 output tokens split across the calls) | ~$0.33 |
- Context size and request size are different things. Crossing 272K reprices the entire request.
- Long conversations benefit from $0.05 cached input — until the request passes 272K, when cache rates double to $0.10/M.
- All figures are arithmetic from official multipliers; check actual billing in API responses.
Specifications (official model page)
Source: OpenAI — GPT-6.1 Terra (retrieved 2026-10-29). Tool calling requires the Responses API.
Benchmarks (OpenAI's own harness, OpenAI-published numbers)
Every figure below is OpenAI's published result on its own evaluation harness, not an independent third-party verification.
| Benchmark | GPT-6.1 Terra | What the post compares it to |
|---|---|---|
| Agents' Last Exam (v1, medium effort) | 15 points above GPT-5.6 Terra | At about one-third the cost; at max effort ~3x more token-efficient than Claude Opus 5.5 and ~9x lower cost per task |
| MMMU-Pro (visual reasoning) | Beats GPT-5.6 Terra at every reasoning setting | "Nearly matches" GPT-6 Astra and GPT-6.1 Sol at much lower cost |
| OSWorld 2.1 (desktop apps) | Tops GPT-5.6 Terra's best score at medium effort | At "a fraction of the cost per task" |
| Terminal-Bench 4.0 (max effort) | 50.3% | More than double GPT-5.6 Terra's published 21.5% |
| Terminal-Bench-Science 0.1 (max effort) | Above GPT-6.1 Sol | Within 5 points of Opus 5.5 at under one-fifth the estimated cost |
| Internal error-inducing prompts (low effort) | 16.7% error rate | Down from 53.6% on GPT-5.6 Terra |
Source: OpenAI — Introducing GPT-6.1 Terra (retrieved 2026-10-29). Third-party scores (e.g. Arena) were not confirmed at publication time.
Availability
- ChatGPT Work and Codex: available to all Plus, Pro, Enterprise and Edu users (official wording).
- API: model ID
gpt-6.1-terra(official post and model page). - Stage: general availability. The post lists no waitlist, preview or research-preview label.
- Third-party: GitHub's changelog announced GPT-6.1 Terra as generally available in GitHub Copilot on 2026-10-26 (treated here as third-party information).
Source: OpenAI — Pricing and availability (retrieved 2026-10-29)
Where it fits, and what to watch
- Built for high-volume work: invoice reading and data entry, form filling, video editing — the everyday tasks the launch post names.
- Not the pick for the hardest coding: OpenAI itself says Sol and Astra "remain the better choice" there.
- Safety: OpenAI states Terra meets the "Critical" cybersecurity threshold under its Preparedness Framework; see the system card.
- Prices and specs change; always re-check the official page before committing.
Related: GPT-6.1 Sol · Pricing comparison · Token calculator · Timeline