GPT-6.1 Terra costs $1 in and $5 out per million tokens, and guesses far less often when it does not know the answer

OpenAI's new mid-tier model is cheaper and admits what it does not know more often. Both matter for automations that run all day.

What happened

OpenAI released GPT-6.1 Terra, a mid-range model aimed at everyday work like drafting proposals, reading technical drawings and filling out forms. API pricing is $1 per million input tokens and $5 per million output tokens, down from $2 and $12 for GPT-5.6 Terra. Cached input drops to $0.05 per million tokens, a quarter of the old rate.

Artificial Analysis says that at maximum reasoning effort Terra gains 6 points on its Intelligence Index over GPT-5.6 Terra at less than a third of the cost per task ($0.40 versus $1.40). On its AA-Omniscience benchmark, Terra guesses 38% of the time when it does not know an answer, down from 88%, while accuracy rose 2 points.

The Decoder reports OpenAI's claim that Terra matches GPT-6 Astra on the Agents' Last Exam computer use benchmark at one tenth of the token price, and notes OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the hardest coding work. Terra is available in the API, in Codex and in ChatGPT for Plus, Pro, Enterprise and Edu users.

My take

In most client automations the model bill comes from volume: classifying leads, summarizing calls, pulling fields from forms, drafting follow ups. Those steps rarely need the top model, but teams often route them to one because the cheaper model made things up. A guess rate falling from 88% to 38% is the number I care about here, more than the price.

How I would act on it:

1. List every AI step in your n8n, Make or Zapier flows with its model and monthly cost.
2. Pick the high-volume steps and run 50 real past inputs through Terra next to your current model.
3. Look hardest at inputs where the right answer is "unknown" or "not in the document". That is where cheap models used to fail.
4. Switch only the steps where results hold, and keep the model name in one config value so you can roll back in a minute.

The $0.05 cached input price also rewards a stable prompt. Put the long instructions first and the changing data last so more of each call hits the cache.

One caution: benchmarks are not your data, and 38% is still guessing. Keep validation on any field that writes to a CRM.