In this article
OpenAI released GPT-6.1 Terra on October 26, 2026 as the successor to GPT-5.6 Terra for visual reasoning and computer-use pipelines. OpenAI lists standard API pricing at $1 per million input tokens and $5 per million output tokens. MarkTechPost's comparison puts that at half GPT-6.1 Sol's $2/$10 and one-tenth GPT-6 Astra's $10/$50. OpenAI also reports Terra matches Astra on Agents' Last Exam v1 at medium reasoning effort at a fraction of the per-task cost. For high-volume screen automation, invoice parsing, or document interpretation, the combination of that parity and a much lower cost floor is the operative change.
Benchmark performance: agents and vision
OpenAI's launch charts report Terra scoring 57.6% on Agents' Last Exam v1 at medium effort, identical to GPT-6 Astra but at $0.29 per task versus Astra's $4.82. At maximum effort, Terra reaches 60.0% at $1.01 per task, while Claude Opus 5.5 scores 63.2% at $8.82 per task — a 3.2-point quality gap for roughly 9x the cost per task.
On OSWorld 2.1 offline desktop tasks, Terra scores 66.7% at medium effort and 74.2% at maximum, against GPT-5.6 Terra's 57.1% peak. That trails GPT-6.1 Sol (78.0%), GPT-6 Astra (80.9%), and Claude Sonnet 5.5 (83.9%). On MMMU-Pro, MarkTechPost reports Terra reaches 85.5% at maximum effort, within 1.2 points of Astra's 86.7% and 0.8 points of Sol's 86.3%, at $0.012 per task versus $0.065 for Astra. GPT-5.6 Terra topped out at 81.7%.
Factuality and coding
OpenAI reports that on its internal adversarial prompt set, the low-effort error rate fell from 53.6% for GPT-5.6 Terra to 16.7% for GPT-6.1 Terra, and to 4.3% at maximum effort — nearly matching GPT-6 Astra and GPT-6.1 Sol. Because the prompt set is adversarial by design, production error rates will differ.
On Terminal-Bench 4.0, Terra solves 50.3% at maximum effort, up from 21.5% for GPT-5.6 Terra but trailing GPT-6.1 Sol (55.9%), GPT-6 Astra (58.2%), and Claude Sonnet 5.5 (70.6%); OpenAI recommends Sol or Astra for the most complex coding tasks. On Terminal-Bench-Science 0.1, OpenAI reports Terra scores 58.3% at $4.15 per task, ahead of GPT-6.1 Sol's 57.0% at $5.47 but behind Claude Opus 5.5 with fallbacks at 63.3% for $23.21.
Pricing and availability
| Model | Input ($/1M tokens) | Output ($/1M tokens) | ALE v1 max | ALE max cost/task | Terminal-Bench 4.0 max | MMMU-Pro max |
|---|---|---|---|---|---|---|
| GPT-6.1 Terra (OpenAI) | $1.00 | $5.00 | 60.0% | $1.01 | 50.3% | 85.5% |
| GPT-6.1 Sol (OpenAI) | $2.00 | $10.00 | 58.6% | $2.00 | 55.9% | 86.3% |
| GPT-6 Astra (OpenAI) | $10.00 | $50.00 | 59.3% | $7.24 | 58.2% | 86.7% |
| Claude Opus 5.5 (Anthropic) | $4.00 | $20.00 | 63.2% | $8.82 | — | — |
| Claude Sonnet 5.5 (Anthropic) | $2.00 | $10.00 | — | — | 70.6% | — |
| Claude Haiku 4.5 (Anthropic) | $1.00 | $5.00 | — | — | — | — |
Benchmark scores and per-task costs are vendor-reported from OpenAI's launch charts as compiled by MarkTechPost. Claude Haiku 4.5 list price is from AI Mastery pricing data (as of 2026-10-22); Claude Opus 5.5 and Claude Sonnet 5.5 list prices are from MarkTechPost's comparison table.
The model is available through the OpenAI API as gpt-6.1-terra and in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu subscribers. Cached input costs $0.05 per million. MarkTechPost reports Batch and Flex modes cut standard pricing by 50%, Fast mode doubles it, and prompts above 272,000 input tokens incur 2x input and 1.5x output surcharges. The context window is 1,050,000 tokens with a 128,000-token output ceiling, and the knowledge cutoff is June 1, 2026. Tool calling — computer use, hosted shell, code interpreter, web search, file search, and MCP — requires the Responses API. Weights are closed; parameter count, architecture, and serving hardware are undisclosed.
AI Mastery analysis
The pricing structure fits the token efficiency repricing pattern we have seen across 2026 releases: a capable mid-tier model priced to make the frontier model economically irrational for high-volume workloads. At $1/$5, Terra occupies the same list-price tier as Claude Haiku 4.5, according to AI Mastery's pricing data.
The benchmark evidence is split. Agents' Last Exam parity with Astra at medium effort is the headline, but OSWorld 2.1 puts Terra 6.7 points behind Astra and 9.7 points behind Claude Sonnet 5.5 at maximum effort. Those two evaluations appear to measure different properties of computer-use capability, a selection risk covered in our analysis of benchmark framing patterns.
The coding gap is explicit in OpenAI's own guidance: 50.3% on Terminal-Bench 4.0 versus Sonnet 5.5's 70.6% is a 20.3-point spread, making Terra a weaker default for code-heavy automation. OpenAI also reports a 29.6% restriction-routing rate in the Respecting Warnings evaluation at maximum effort, which production deployments should review for scope-limiting controls. Whether OSWorld 2.1 or Agents' Last Exam better predicts real deployment outcomes is now the central selection question for teams choosing between Terra, Sol, and Astra.
Sources
- Primary: Introducing GPT-6.1 Terra | OpenAI
- Reported by: MarkTechPost: OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work
- Competitor prices: AI Mastery token calculator, figures as of 2026-10-22
Frequently asked questions
How much does GPT-6.1 Terra cost per million tokens?
OpenAI lists standard pricing at $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million. MarkTechPost reports Batch and Flex modes cut the standard rate by 50%, and Fast mode doubles it. That is half GPT-6.1 Sol's $2/$10 according to MarkTechPost's comparison.
Does GPT-6.1 Terra match GPT-6 Astra on agent tasks?
OpenAI reports both models score 57.6% on Agents' Last Exam v1 at medium reasoning effort, but Terra's cost is $0.29 per task versus Astra's $4.82. At maximum effort Terra reaches 60.0% at $1.01 per task, while Claude Opus 5.5 scores 63.2% at $8.82.
Does GPT-6.1 Terra run locally or on a single GPU?
No. Terra is an OpenAI-hosted API model and is not available for local deployment. Weights are closed, and OpenAI has not disclosed parameter count, architecture, or serving hardware.
How much did coding improve over GPT-5.6 Terra?
OpenAI reports Terminal-Bench 4.0 performance rose from 21.5% to 50.3% at maximum effort. Terra still trails GPT-6.1 Sol at 55.9%, GPT-6 Astra at 58.2%, and Claude Sonnet 5.5 at 70.6%, and OpenAI recommends Sol or Astra for the most complex coding.
What are GPT-6.1 Terra's context and output limits?
The context window is 1,050,000 tokens, with a maximum output of 128,000 tokens. Its knowledge cutoff is June 1, 2026.
Related Reading
GPT-6.1 Sol Matches Astra on Coding at One-Fifth the Price
OpenAI's GPT-6.1 Sol hits $2/$10 per million tokens—one-fifth of Astra's rate—while matching it on DeepSWE and closing to within 2.1 points on computer use.
GPT-6 Sol and Luna Cut API Prices 50%—Benchmarks Included
OpenAI's GPT-6 Sol ($2/$10 per 1M tokens) and Luna ($0.10/$0.50) go live in the API with benchmark results against Claude Opus 5 and Fable 5.
GPT-6 Astra: $10/M Tokens, 57.9% Terminal-Bench, Critical Cyber Flag
OpenAI's GPT-6 Astra scores 57.9% on Terminal-Bench 4.0, costs $10/$50 per million tokens, and is the first model to hit the Critical cybersecurity threshold.