Reviews · OCTOBER 28, 2026
GPT-6.1 Terra lands at $1/$5, cutting computer-use token costs tenfold vs. Astra
OpenAI's October 26 release matches GPT-6 Astra on Agents' Last Exam at a tenth the per-token price, scores 50.3% on Terminal-Bench 4.0 at max effort, and runs tasks over three times faster than Sol and Astra.
OpenAI released GPT-6.1 Terra on October 26 at $1 per million input tokens and $5 per million output, half the API price of both GPT-5.6 Terra and GPT-6.1 Sol, and a tenth of what GPT-6 Astra charges per token for comparable computer-use work. Cached input drops to $0.05 per million. The pricing is the story; the benchmarks explain why it matters.
Terra scores 60.0% on Agents' Last Exam, which OpenAI frames as a 15-point jump over GPT-5.6 Terra at roughly one-third the cost, and a match for Astra at one-tenth Astra's per-token rate. On Terminal-Bench 4.0 at maximum reasoning effort it posts 50.3%, more than double GPT-5.6 Terra's 21.5%. On Terminal-Bench-Science 0.1 it lands at 58.3%, within six points of Astra's 64.6%, and it hits 74.2% on the OSWorld 2.1 offline subset. "Terra is also much quicker, completing tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol," OpenAI said in the launch post.
The factuality numbers are the quietly impressive part. On OpenAI's internal error-inducing-prompt benchmark, GPT-5.6 Terra at low reasoning produced errors 53.6% of the time. Terra posts 16.7% at default and 4.3% at max effort. That curve, more than the headline Agents' Last Exam score, is what makes continuous-operation agents plausible for teams that can't babysit every run.
Against Claude Opus 5.5, OpenAI claims Terra is over three times as token-efficient at max reasoning, a quarter of the per-token price, and roughly 9x cheaper per completed task. The Next Web notes those figures haven't been independently verified.
For small B2B operators, the pricing floor is where theory meets payroll. Agentic workflows that read screens, enter invoice data, qualify leads, and fill CRM forms were priced for enterprises when the only option was Astra-class compute. Terra brings them into the range where a two-person team can run them continuously, following a pattern already visible in GPT-6.1 Sol's one-fifth-of-Astra launch pricing and Claude Haiku 5.5's $0.10/M price cut.
Context is worth naming. Terra ships weeks after OpenAI cancelled GPT-6.1 Astra over failed internal safety tests, per The Next Web. OpenAI classifies Terra at the Critical cybersecurity threshold under its Preparedness Framework and says it carries forward Astra's protections. The model is available as gpt-6.1-terra in the API and inside ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users. BleepingComputer notes OpenAI still recommends Sol or Astra for the hardest coding work.
Mid-tier is where the economics of agents actually get decided. Terra is the clearest signal yet that the frontier labs understand this.
Sources
- https://openai.com/index/introducing-gpt-6-1-terra/
- https://openai.com/products/release-notes/
- https://benchlm.ai/models/gpt-6-1-terra
- https://thenextweb.com/news/openai-gpt-6-1-terra-computer-use
- https://www.bleepingcomputer.com/news/artificial-intelligence/openai-launches-gpt-61-terra-with-3x-faster-ai-agents-50-percent-lower-pricing/