← Back to feed

OpenAI shipped GPT-6.1 Terra on October 26, 2026, as gpt-6.1-terra on the API and in ChatGPT Work and Codex. Standard pricing: $1 per million input tokens, $5 per million output tokens, $0.05 per million cached input tokens. That is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol.

Benchmark results

On Terminal-Bench 4.0, GPT-6.1 Terra scores 50.3% at maximum reasoning effort, compared to GPT-5.6 Terra’s published score of 21.5% — more than double. The gap narrows against GPT-6.1 Sol and GPT-6 Astra at max effort, where both remain ahead for the most complex coding tasks.

Artificial Analysis placed GPT-6.1 Terra at 48 on its Coding Agent Index v2.0, below GPT-6.1 Sol (54) and above GPT-6 Luna (31). At low and medium effort settings, the model sits on the Pareto frontier for cost efficiency in that index.

On Agents’ Last Exam — a benchmark of professional computer-use tasks including turning architectural drawings into 3D models and creating time-coded restoration logs from animated footage — GPT-6.1 Terra at medium reasoning effort scores 15 percentage points above GPT-5.6 Terra at roughly one-third the cost. It matches GPT-6 Astra at one-tenth the price per token. Task completion runs over three times faster than both GPT-6 Astra and GPT-6.1 Sol.

Terminal-Bench-Science 0.1, which tests code-based scientific research tasks, shows GPT-6.1 Terra surpassing GPT-6.1 Sol at lower cost at max reasoning, coming within five points of Claude Opus 5.5 at less than one-fifth the estimated cost.

Hallucination drop

On AA-Omniscience, Artificial Analysis’ private knowledge and hallucination benchmark, GPT-6.1 Terra guesses incorrectly on 38% of questions where it does not know the answer — down from 88% for GPT-5.6 Terra, a 50-point reduction. AA-Omniscience accuracy also rises by 2 points, meaning the model answers more questions correctly while guessing less.

OpenAI’s internal error-rate benchmark shows a sharper improvement: at low reasoning effort, the error rate fell from 53.6% (GPT-5.6 Terra) to 16.7%, a 37-percentage-point drop. At max effort, it reaches 4.3%, near the rates for GPT-6 Astra and GPT-6.1 Sol.

Computer use and visual reasoning

MMMU-Pro results show GPT-6.1 Terra outperforming GPT-5.6 Terra at every reasoning setting, nearly matching GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, at medium reasoning effort, Terra surpasses GPT-5.6 Terra’s best score. The model targets desktop-based multi-step tasks: building 3D models from technical drawings, reviewing footage frame by frame, reading invoices and entering data into accounting software.

Positioning and context

The model carries a 1.05 million token context window. OpenAI classifies it as Critical in cybersecurity under its Preparedness Framework and applies the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra. It is not available through UAE in-region routing.

Within the GPT-6.1 family, the positioning is explicit: GPT-6.1 Sol and GPT-6 Astra for complex coding and research, Terra for high-volume computer-use and professional workday tasks where cost efficiency matters. With GPT-6.1 Sol priced at $2/$10 per million tokens, Terra’s $1/$5 rate makes it the lower-cost routing option for code and agentic workflows that do not need Sol-class reasoning depth.