2 sources 1m ago
OpenAI Releases GPT-6.1 Terra as a Cheaper, Faster Model
OpenAI released GPT-6.1 Terra on October 27, 2026, Japan time. The model is a major update to GPT-5.6 Terra, released as 6.1 while skipping 6.0.
Reporting from 2 sources: ASCII.jp, GIGAZINE.
Why it matters
On OpenAI's OSWorld 2.1 PC operation test, GPT-6.1 Terra matched GPT-6 Astra's score at lower cost while running three times faster, shifting the cost-performance tradeoff for agent work.
Key Facts
- OpenAI released GPT-6.1 Terra on October 27, 2026, Japan time.
- GPT-6.1 Terra API pricing is $1 per million input tokens and $5 per million output tokens, half the standard price of GPT-6.1 Sol, according to ASCII.jp.
- GPT-6.1 Terra was trained on data up to June 1, 2026, and has a context window of 1.05 million tokens with maximum output of 128,000 tokens.
- GPT-6.1 Terra reached a 50.3% success rate on Terminal-Bench 4.0 at maximum reasoning, against 21.5% for GPT-5.6 Terra.
- Artificial Analysis scored GPT-6.1 Terra at 48 on its Intelligence Index v4.3.2, within 4 points of GPT-6.1 Sol.
OpenAI offers four model tiers: the flagship Astra, high-performance Sol, balanced Terra, and low-cost Luna. GPT-6.1 Terra is available via API and to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. API pricing is $1 per million input tokens, $0.05 cached input, and $5 output, held to half the standard price of GPT-6.1 Sol, according to ASCII.jp. The model was trained on data up to June 1, 2026, with a context window of 1.05 million tokens and maximum output of 128,000 tokens. Reasoning depth has six levels: none, low, medium, high, xhigh, and max. OpenAI says GPT-6.1 Terra handles image understanding and PC operation at lower cost than higher-end models.
On the Agents' Last Exam benchmark, which includes tasks such as building 3D models from blueprints, GPT-6.1 Terra scored 15 points higher than GPT-5.6 Terra at medium reasoning, and cost per task was about one-third of the previous model's. On Terminal-Bench 4.0, which tests problem solving through command operations, it reached a 50.3% success rate at the maximum reasoning setting, up from the 21.5% published figure for the older Terra. In an internal evaluation using prompts designed to induce factual errors, the error rate at weak reasoning fell from 53.6% to 16.7%; at maximum reasoning it was 4.3%.
Third-party group Artificial Analysis scored the model 48 on its Intelligence Index v4.3.2, within 4 points of GPT-6.1 Sol. On MMMU-Pro(Vision), GPT-6.1 Terra reached a comparable score to GPT-6 Astra at lower cost. On Agents' Last Exam it was comparable to Astra at one-tenth the cost. OpenAI employee Victor E. Nunez said "when GPT-6.1 Terra and GPT-6.1 Sol were made to fill out a web page form, GPT-6.1 Terra completed the task more than 2 minutes faster." CEO Sam Altman also cited the model's PC operation performance. OpenAI notes Sol and Astra remain better suited to the most complex coding work. A safety system card for Terra was published as an addendum to the GPT-6 Astra System Card.
Written with AI by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google