OpenAI launches GPT-6.1 Terra for visual reasoning, computer use and writing

What OpenAI released

OpenAI has published API documentation for a new model, GPT-6.1 Terra, which reached the front page of Hacker News on 26 October. According to OpenAI's model reference, Terra is positioned as a quick, resource-efficient model built around three headline strengths: reasoning over images, accurate control of computers, and natural written prose.

The documentation gives concrete examples of intended use: pairing product photos with catalog records, interpreting technical drawings, navigating and acting within application screens, and producing or editing text. That mix places Terra in the "agent that can see and act" category rather than the pure chat lane.

Reasoning control and tooling

Developers can tune how much reasoning the model applies via a reasoning.effort setting with six levels: none, low, medium (the default), high, xhigh and max. The wide range gives teams a dial for trading answer quality against latency and token cost, from skipping deliberation entirely to maximum-effort reasoning.

Tool calling is available, but only through OpenAI's Responses API. When used there, the model supports built-in image generation and code interpreter tools. Function calling and structured outputs are both listed as supported features.

The model supports up to 128,000 output tokens and has a knowledge cutoff of 1 June 2026. OpenAI also points developers to a separate GPT-6 guide for prompting and migration advice.

Endpoints and pricing

Terra is exposed across a broad set of endpoints, including Chat Completions, fine-tuning, image generation and editing, speech generation, transcription, translation, realtime translation and realtime transcription, plus the legacy completions API.

Pricing follows the usual per-token structure with several multipliers:

• Cached input tokens are billed at 5% of the uncached input rate.
• Cache writes cost $1.25 per million tokens for 30-minute retention and $2.50 for 12-hour retention, before long-context adjustments.
• Requests exceeding 272,000 input tokens are charged at double the input and cache rates and 1.5x the output rate for the entire request.
• Batch and Flex processing are 50% cheaper than Standard.
• Regional processing, where offered, adds a 10% premium.
• Fast mode is priced at 2x Standard.

For tool-specific workloads such as search and computer use, OpenAI charges a fee per tool call on top of token costs.

Rate limits

The free usage tier does not support the model at all. Paid tiers scale from there: Build allows 5,000 requests per minute and one million tokens per minute, Launch raises that to 10,000 requests and four million tokens per minute, and Grow reaches 15,000 requests and 40 million tokens per minute. Batch queue limits climb accordingly, from three million tokens at Build to 15 billion at Grow. Limits increase automatically as accounts send more requests and spend more through the API.

Why it matters

The release signals where OpenAI believes developer demand is heading. Folding computer use, visual reasoning and image tools into the model itself, rather than treating them as bolt-ons, reflects a bet that agents acting inside real applications are becoming the default workload. The six-level reasoning dial suggests that cost control over reasoning tokens is now a first-class product concern, not an afterthought.

The pricing structure rewards careful architecture: the surcharge on requests above 272K input tokens and the per-tool-call fees mean teams building long-context or agent-heavy systems will need to model their costs deliberately. And with no free-tier access, experimenting with Terra requires a paid plan from the start, raising the barrier for smaller developers evaluating whether it beats their current model.