GPT-6.1 Terra: $1 input, $5 output for OpenAI's mid-tier model

GPT-6.1 Terra is OpenAI’s new mid-tier model, and the headline is price: $1 per million input tokens and $5 per million output tokens through the API. That is half the input price and well under half the output price of GPT-5.6 Terra, which it replaces. OpenAI pitches it for visual reasoning, computer control and everyday office work. The Decoder reported it on October 26, 2026, and Artificial Analysis published its own measurements.

What changed

The price cut is the clearest change:

Per 1M tokens	GPT-5.6 Terra	GPT-6.1 Terra	Claude Sonnet 5.5
Input	$2	$1	$2
Output	$12	$5	$10
Cached input	$0.20	$0.05	not compared

Artificial Analysis says writing to the cache costs $1.25 per million tokens, also half the old rate.

Quality moved up at the same time. On the Artificial Analysis Intelligence Index, Terra at maximum reasoning effort scores 6 points above GPT-5.6 Terra and 4 points below GPT-6.1 Sol, OpenAI’s larger model. At lower effort settings, the gain over the old Terra is 7 to 13 points. Artificial Analysis estimates the cost per task at about $0.40, against $1.40 for GPT-5.6 Terra.

Other reported results:

• Coding Agent Index v2.0: 48, against 54 for GPT-6.1 Sol and 31 for GPT-6 Luna.
• Hallucination: on AA-Omniscience, Terra guesses when unsure 38% of the time, down from 88% for GPT-5.6 Terra. Accuracy rose 2 points.
• Terminal-Bench 4.0: 50.3% at max effort, up from 21.5%, according to The Decoder.
• Prompt injection: OpenAI reports 98.8% defense against indirect prompt injection, up from 96.2%.

OpenAI rates Terra at its “Critical” cybersecurity level under its Preparedness Framework and says it carries over the safeguards of GPT-6 Astra.

Why it matters

Mid-tier models do most of the work in real products: support bots, document handling, agents that click through software. A price cut here changes more bills than a cut to a flagship.

At $1 and $5, Terra undercuts Claude Sonnet 5.5, its most direct rival, on both input and output, according to The Decoder’s comparison. The drop in hallucinated guesses may matter even more than the price. A model that admits it does not know is cheaper to run safely, because fewer answers need checking.

It also shows how fast the middle of the market moves. This month we have already covered the Claude Haiku 5.5 cuts and the Sol and Opus 5.5 cuts. Prices set in September are already out of date.

OpenAI still recommends Sol and Astra for the hardest coding work, so Terra is not a replacement for everything. The 6-point gap to Sol on the coding agent index is real.

What to test

• Your real tasks at your usual effort setting. The gains differ by reasoning effort. Test at the setting you would actually pay for.
• Cost per finished task, not per token. Higher effort uses more tokens. Compare total cost for a batch of real jobs against your current model.
• Hallucination on your data. The 38% figure comes from one benchmark. Ask questions your documents cannot answer and see whether Terra says so.
• Computer use. OpenAI highlights computer control, but several results are given only as relative comparisons without exact scores. Measure completion rates yourself.
• Limits. Neither report gives the context window or rate limits. Check these in OpenAI’s documentation before you plan a migration.

Common questions

How much does GPT-6.1 Terra cost?

$1 per million input tokens, $5 per million output tokens, and $0.05 per million cached input tokens, according to The Decoder and Artificial Analysis.

Is GPT-6.1 Terra better than GPT-6.1 Sol?

No. Artificial Analysis puts Terra 4 points below Sol on its Intelligence Index and 6 points below on its coding agent index. Terra is the cheaper option for everyday work, and OpenAI still recommends Sol and Astra for the hardest coding tasks.

Where can I use GPT-6.1 Terra?

In ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API.

The conclusion

GPT-6.1 Terra is the kind of release that should change a spreadsheet: lower prices, better scores and far fewer confident guesses than the model it replaces. The comparisons come from OpenAI and Artificial Analysis, and some results lack exact scores, so run your own tests before moving production work. Watch two things next: how Anthropic prices the next Sonnet, and whether Google’s Gemini 4 line gets a cheaper mid-tier option.