OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work
By Michal Sutter — October 26, 2026
OpenAI has released GPT-6.1 Terra, a fast model for visual reasoning, computer use, and writing. It replaces GPT-5.6 Terra and is live in the API as gpt-6.1-terra. Standard pricing is $1 per million input tokens and $5 per million output tokens—roughly half the cost of GPT-6.1 Sol. For teams automating screens, forms, and documents at volume, GPT-6.1 Terra meaningfully changes the cost math.
TL;DR
- Size: 1.05M-token context, 128K max output.
- Runs on: OpenAI-hosted only (API, ChatGPT Work, Codex). Closed weights.
- Performance: Near GPT-6.1 Sol and GPT-6 Astra on vision and computer use, at lower cost.
- Best: 85.5% on MMMU-Pro (Vision) at max effort, 1.2 points behind GPT-6 Astra.
- Worst: 50.3% on Terminal-Bench 4.0, behind GPT-6.1 Sol (55.9%) and Claude Sonnet 5.5 (70.6%).
What Is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI reasoning model built for high-volume visual and computer-use work. It accepts multimodal inputs—in particular screenshots and rendered documents—and is positioned between OpenAI’s flagship reasoning models and its cheaper, lighter-weight tiers.
With a 1.05M-token context window and 128K maximum output, Terra is designed for long-running agent workflows: multi-page document review, form filling, UI navigation, and iterative writing tasks where the model must retain state across many steps.
The model is available exclusively through OpenAI-hosted surfaces—the API, ChatGPT Work, and Codex—with closed weights, consistent with OpenAI’s broader 2026 strategy of keeping frontier and near-frontier models proprietary while monetizing them through volume-based API pricing.
Positioning Against GPT-6.1 Sol and GPT-6 Astra
Terra’s release lands in a crowded 2026 model lineup. GPT-6 Astra remains OpenAI’s top-end reasoning model, while GPT-6.1 Sol is the established workhorse for agentic and coding tasks. Terra slots in below Sol on price while closing much of the gap on vision benchmarks.
That trade-off is deliberate. Terra performs closest to the frontier on perceptual tasks—reading interfaces, interpreting charts, parsing scanned forms—and furthest from it on execution-heavy terminal and shell tasks, where Sol and competing models from Anthropic retain an edge.
Benchmark Snapshot
| Benchmark | GPT-6.1 Terra | Comparison |
|---|---|---|
| MMMU-Pro (Vision), max effort | 85.5% | 1.2 pts behind GPT-6 Astra |
| Terminal-Bench 4.0 | 50.3% | Behind GPT-6.1 Sol (55.9%) and Claude Sonnet 5.5 (70.6%) |
What It Means for Agent Builders
The headline for engineering teams is cost efficiency at scale. At $1/$5 per million tokens, Terra undercuts Sol by roughly half on standard pricing, which matters for workloads that burn millions of tokens per day on screenshot analysis, form automation, and document processing.
In practice, that suggests a tiered architecture: route high-volume perceptual and writing subtasks to Terra, and reserve Sol or Astra for terminal execution, complex multi-step reasoning, and cases where a few benchmark points translate into materially better reliability.
Terra is available now via the API as gpt-6.1-terra, alongside ChatGPT Work and Codex.
via MarkTechPost