OpenAI releases GPT-6.1 Terra, at half GPT-6.1 Sol's price on every line of the pricing table

OpenAI released GPT-6.1 Terra, model id gpt-6.1-terra, on 26 October 2026: a fast, efficient model for visual reasoning, precise computer use and clear, natural writing, on the Responses and Chat Completions endpoints. On the pricing page it costs half of GPT-6.1 Sol on every column — $1.00 input and $5.00 output per million tokens, against Sol's $2.00 and $10.00. If you run image work, screen automation or writing ongpt-6.1-sol or gpt-5.6-terra, measure it.

What changed

The model page prices text tokens, per 1M, at $1.00 input, $0.05 cached input, $1.25 cache writes and $5.00 output. Cached input costs 5% of the uncached rate. The $1.25 write is for 30-minute retention; a 12-hour write costs $2.50, both before long-context adjustments. Past 272K input tokens the whole request is repriced — 2x the input and cache rates and 1.5x output. Batch and Flex run 50% below Standard, regional processing adds 10% where it is offered, and Fast mode costs 2x Standard.

The pricing page's flagship table — input, output, cached input, 30-minute write, 12-hour write — now has four rows:gpt-6-astra at $10.00, $50.00, $1.00, $12.50 and $25.00; gpt-6.1-sol at $2.00, $10.00, $0.10, $2.50 and $5.00; gpt-6.1-terra at $1.00, $5.00, $0.05, $1.25 and $2.50; gpt-6-lunaat $0.10, $0.50, $0.01, $0.125 and $0.25. The model catalogue lists it as GPT-6.1 Terra, model idgpt-6.1-terra.

The model page names the work it is meant for: matching product photos to catalog entries, interpreting technical drawings, acting on application screens, or drafting and revising text. It lists a 1,050,000-token context window, 128,000 maximum output tokens and a knowledge cutoff of 1 June 2026. It takes text and images in and gives text out; audio and video are not supported. Streaming, function calling and structured outputs are supported; fine-tuning is not. Through the Responses API it supports web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search.

Two details decide whether your calls carry over. reasoning.effortaccepts none, low, medium (the default), high, xhigh and max — unlike GPT-6.1 Sol and GPT-6 Astra, which do not supportnone. And tool calling requires the Responses API.

What it was before

The previous Terra was GPT-5.6 Terra, released on 9 July as the GPT-5.6 model "for a balance of intelligence and cost". From 30 July it cost 20% less. In the GPT-6 family that balancing role belongs to GPT-6.1 Sol , which we covered inOpenAI releases GPT-6.1 Sol; the GPT-6 guide calls Terra "fast and efficient", between Sol's balance and Luna's lowest cost. As that item and our item on 12-hour caching recorded, the flagship table then had three rows; Terra is the fourth, and there is nogpt-5.6-terrarow. So set Terra's prices against what your own bill shows forgpt-5.6-terra; for whether and when that model stops being served, the changelog points to the deprecations page.

What it means for you

If you are on gpt-6.1-sol, Terra halves every line of your bill: input from $2.00 to $1.00, output from $10.00 to $5.00, cached input from $0.10 to $0.05, and both cache writes. What you give up is not stated as a number: the guide places Sol as balanced and Terra as fast and efficient. If your traffic is photos, drawings, screens or prose — the page's own examples — run your evaluation set on both before moving production traffic.

If you are on gpt-5.6-terra, the guide's advice is to compare both models on representative images and workflows using the same reasoning effort. If your application calls tools through Chat Completions, those requests move to the Responses API.

If you are on gpt-6-luna, Terra is not a saving: it costs ten times as much on input ($1.00 against $0.10) and output ($5.00 against $0.50). And the catalogue still sends cost-sensitive, high-volume work to Luna.

Before you switch, check five things.

• The id string. Send gpt-6.1-terra— the only snapshot the model page lists. -Your reasoning setting. none works here, so a call moved from gpt-6-sol or gpt-6-luna with nonekeeps it. A request sendingminimalshould start at low. The guide's default advice is medium, raised only when the quality gain is worth the time and tokens.
• Your tool calls.Tool calling requires the Responses API.
• Your cache writes.Use the 30-minute price ($1.25) or the 12-hour price ($2.50) that matches your retention ; our item on12-hour prompt caching explains the choice.
• Your rate limits.Requests per minute, tokens per minute and batch queue follow your usage tier : 5,000, 1,000,000 and 3,000,000 at Build; 10,000, 4,000,000 and 200,000,000 at Launch; 15,000, 40,000,000 and 15,000,000,000 at Grow. The Free tier is not supported.

If you need Fast mode with EU data residency, Terra has it; GPT-6 Astra and GPT-6 Sol do not.

The source

OpenAI's API changelog, the entry of 26 October 2026, with the GPT-6.1 Terra model page, the model catalogue, the pricing page and the GPT-6 guide.