Error rates on adversarial factual questions at low effort dropped from 53.6% to 16.7%. OpenAI has released the GPT-6.1 Terra model, priced via API at half the cost of GPT-5.6 Terra and GPT-6.1 Sol: $1 per million input tokens, $5 per million output tokens, and $0.05 for cache hits. This new API machine undercuts previous versions by 50%, reshaping inference economics and forcing the market to recalculate routine automation expenses.
In Agents' Last Exam professional computer task benchmarks, OpenAI reports a 15 percentage point performance increase over GPT-5.6 Terra at one-third of the cost under medium effort settings. Terra accomplishes tasks at roughly one-ninth the cost of Claude Opus 5.5 while maintaining comparable benchmark scores. At maximum effort on Terminal-Bench 4.0, the new system scores 50.3% compared to 21.5% for the previous Terra version. For complex code generation, OpenAI recommends Sol or Astra, delegating stream-based tasks like invoice parsing, form filling, and desktop UI clicking to Terra.
According to OpenAI's applied AI guide published on August 13, 2026, the GPT-5.6 family fundamentally shifts this paradigm by delivering higher accuracy at reduced inference costs, enabling cost-efficient models to handle complex agentic scenarios.
The model is now available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu tiers, and is designated as gpt-6.1-terra in the API. While reduced inference costs allow organizations to migrate routine workflows to cheaper hardware, developers and enterprise buyers must evaluate accuracy trade-offs in desktop automation and verify whether autonomous agent stability holds across the new architecture.
Editorial analysis
Inference used to demand heavy budgets, but deploying GPT-6.1 Terra now costs just $1 per million input tokens and $5 per million output tokens, with caching at $0.05. As we noted in late August, the GPT-5.6 family cut reasoning costs for complex agents, yet this new release slashes API prices by half compared to both GPT-5.6 Terra and GPT-6.1 Sol. Will the broader ecosystem maintain service margins amid such rapid drops in compute costs — the release leaves unexplained.
We covered this: March 25, August 24