GPT-6.1 Terra
- ID
- 39718
- Status
- summarized
- Published
- 27 Oct 2026, 2:01 AM
- Fetched
- 27 Oct 2026, 11:14 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://openai.com/index/introducing-gpt-6-1-terra/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.0
- Created
- 27 Oct 2026, 11:15 AM
- Tags
- Audience
- developersai_ml_learnersai_agent_userssaas_foundersvibe_coders
What happened
OpenAI announced GPT-6.1 Terra, a faster, cheaper model aimed at computer use and visual reasoning, with standard API prices 50% below GPT-5.6 Terra and GPT-6.1 Sol. OpenAI reports its internal error-inducing-prompt benchmark dropping from 53.6% error at low reasoning effort (GPT-5.6 Terra) to 16.7%, and 4.3% at max effort, plus a 15-percentage-point gain on Agents' Last Exam at medium effort for about one-third the cost. At max effort OpenAI claims Terra is over three times as token-efficient as Claude Opus 5.5 at one-fourth the price per token, a ~9x lower cost per task, and over three times faster than GPT-6 Astra and GPT-6.1 Sol.
Why it matters
The decision-relevant number is the 50% cut on standard API pricing versus GPT-5.6 Terra and GPT-6.1 Sol — if you run high-volume extraction work like reading invoices into accounting software, your per-document bill changes materially, and the ChatGPT Work/Codex allowance covering more tasks per plan shifts subscription math too. Every benchmark here (Agents' Last Exam, MMMU-Pro, OSWorld 2.1, the 53.6%→16.7% error rate) is self-reported by the vendor and the pasted text is truncated mid-sentence, so treat the claims as a hypothesis to test on your own eval set before migrating production traffic, and compare cost per completed task rather than per-token list price.
Discussion angle
Run a 30-minute bake-off: take 20 real invoice or form documents, run them through your current model and Terra, and compare cost per correctly completed task — then ask whether the 50% list-price cut survives once retries and max-effort calls are counted.