- GPT-6.1 Terra is OpenAI’s new mid-tier model for computer use, available now in ChatGPT Work, Codex and the API as gpt-6.1-terra at $1 and $5 per million tokens.
- OpenAI says it matches GPT-6 Astra on its Agents’ Last Exam benchmark at one-tenth the price per token and runs tasks more than three times faster than Astra or Sol.
- OpenAI rates Terra Critical for cybersecurity under its Preparedness Framework, so the $5 model ships with the safeguards built for its $50 flagship, GPT-6 Astra.
OpenAI’s newest model is built for the dull, high-volume work that AI agents increasingly do on a screen, and it costs half as much as the model it sits beneath. GPT-6.1 Terra, released on October 26, reads interfaces, clicks through apps and fills in forms, and lists at $1 per million input tokens and $5 per million output tokens through the API, according to OpenAI’s launch post. That is 50% below both GPT-6.1 Sol and the GPT-5.6 Terra it replaces.
GPT-6.1 Terra targets screen-based agent work at half the price
Terra fills the gap in the four-tier lineup of Luna, Terra, Sol and Astra that OpenAI left open when it refreshed Sol and Luna in September. The company pitches it at tasks such as reading incoming invoices and entering them into accounting software, interpreting technical drawings and editing video, the kind of repetitive agent work where cost per task decides whether automation pays.
OpenAI’s headline numbers are large. At medium reasoning effort Terra scores 15 points higher than GPT-5.6 Terra on Agents’ Last Exam at about one-third the cost, and at maximum effort OpenAI puts its cost per task at roughly one-ninth of Claude Opus 5.5’s. Its internal error rate on prompts designed to induce mistakes fell from 53.6% to 16.7% at low effort, and its Terminal-Bench 4.0 coding score more than doubled to 50.3%.
| API price per million tokens | $1 input, $0.05 cached input, $5 output |
| Error rate on error-inducing prompts (low effort) | 16.7%, down from 53.6% for GPT-5.6 Terra |
| Terminal-Bench 4.0 (max effort) | 50.3%, up from 21.5% |
| Task speed | Over 3x faster than GPT-6 Astra and GPT-6.1 Sol |
Independent tests confirm the cost gain and a smaller capability jump
Outside measurement supports the efficiency claim and trims the capability story. Artificial Analysis found Terra costs $0.40 per task on its Intelligence Index at maximum effort against $1.40 for GPT-5.6 Terra, and scores 48 on its Coding Agent Index, six points behind GPT-6.1 Sol. Its hallucination rate on the firm’s AA-Omniscience test fell from 88% to 38%, while accuracy rose only two points, so most of the gain comes from the model declining to answer when it does not know.
A cheaper model now carries a Critical cyber rating
The safety disclosure deserves the closest read. OpenAI says Terra meets the Critical threshold for cybersecurity under its Preparedness Framework and keeps the protections built for GPT-6 Astra, which means capabilities once confined to a $50-per-million-token flagship now ship at $5. The pricing extends the cuts OpenAI and Anthropic made in September’s same-day price war, and pushes the mid tier further toward the cost of running agents at scale.
GPT-6.1 Terra makes screen-reading agents cheap enough for routine back-office work, and the Critical cyber rating means every company deploying it at that volume inherits the same misuse risk OpenAI attached to its most expensive model only seven weeks ago.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.