GPT-6.1 Terra: a cheaper API, with a few limits

· 6 min read

#ai

On October 26, 2026, OpenAI released GPT-6.1 Terra, a model it calls fast and efficient, built for visual reasoning, precise computer use and clear writing. The headline is the price: OpenAI says its standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.

In short

The numbers side by side

Prices are per million tokens, for prompts up to 272K input tokens. Sol is in the table because OpenAI compares against it too.

GPT-5.6 Terra GPT-6.1 Sol GPT-6.1 Terra
Input $2 $2 $1
Cached input $0.20 $0.10 $0.05
Output $12 $10 $5
Context window 1,050,000 tokens 1,050,000 tokens 1,050,000 tokens
Maximum output 128,000 tokens 128,000 tokens 128,000 tokens
Knowledge cutoff Feb 16, 2026 Apr 30, 2026 Jun 1, 2026
none reasoning effort supported not supported supported

Terra is the cheapest of the three on every price line. By my arithmetic the output price falls by more than half against GPT-5.6 Terra; that is my calculation, not OpenAI's wording. The window and the output limit did not move. One practical difference from Sol: Terra accepts none as a reasoning effort, so you can switch reasoning off.

The pricing page also lists higher rates for long prompts. Above 272K input tokens, input and cache rates double and output rates rise by half, and the higher rate applies to the whole request. Check where your longest prompts fall.

Where it sits in the timeline

OpenAI's model line has moved fast this year. These are the dates from its API changelog and model pages:

Three OpenAI models and their standard API prices Three boxes from left to right. GPT-5.6 Terra, released July 9, 2026, costs 2 dollars for input and 12 dollars for output per million tokens, after a 20 percent cut on July 30. GPT-6.1 Sol, released September 29, 2026, costs 2 dollars and 10 dollars and is meant for complex coding. GPT-6.1 Terra, released October 26, 2026, costs 1 dollar and 5 dollars. Jul 9, 2026 GPT-5.6 Terra $2 in / $12 out 20% cut on Jul 30 Sep 29, 2026 GPT-6.1 Sol $2 in / $10 out for complex coding Oct 26, 2026 GPT-6.1 Terra $1 in / $5 out 50% lower, says OpenAI Price per 1M tokens, standard rates, input / output
Standard API prices of GPT-5.6 Terra, GPT-6.1 Sol and GPT-6.1 Terra, in the order they launched. From OpenAI's pricing page and changelog.

So the Terra line has got cheaper twice since July: a price cut on GPT-5.6 Terra, and now a new model at a lower price. OpenAI's latest-model guide splits the jobs: Terra for visual and writing work, Sol for complex coding, computer use and professional work. The announcement also names computer use as a Terra strength, so the line between the two is not sharp.

What OpenAI claims about quality

The announcement calls GPT-6.1 Terra a significant step up from GPT-5.6 Terra. On Agents' Last Exam at medium effort it scores 15 points higher at about a third of the cost. It finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol, and it beats GPT-5.6 Terra on MMMU-Pro at every effort setting. Two numbers stand out:

OpenAI's own tests GPT-5.6 Terra GPT-6.1 Terra
Terminal-Bench 4.0, max reasoning, tasks solved 21.5% (published score) 50.3%
Error rate on error-prone prompts, low effort (internal) 53.6% 16.7%

All of these are OpenAI's results, and the error-rate row comes from an internal benchmark. I read them as a reason to test, not as a result for your workload.

The good and the not-so-good

What works better

What is missing, unclear or costs something

My take, not fact: for high-volume jobs like OpenAI's own example of reading incoming invoices into accounting software, the price drop probably matters more than the benchmark gains. I would still run my own sample first.

Switching from GPT-5.6 Terra: a checklist

This follows OpenAI's migration advice in its latest-model guide.

  1. Compare at the same effort. Run both models on representative images and workflows from your own app, and keep your current reasoning effort. Start at medium and raise it only if the quality gain is worth the extra time and tokens.
  2. Move tool calling to the Responses API if you use tools through Chat Completions.
  3. Remove sampling parameters. When reasoning is on (anything other than none), remove temperature, top_p and top_logprobs, and logprobs as well in Chat Completions.
  4. Check your longest prompts against the 272K-token threshold before you trust the new price.

The guide has no Terra-specific prompting advice, so there is nothing more to follow there.

Sources