OpenAI has launched GPT‑6.1 Terra, a model designed for speed, cost efficiency, improved visual reasoning, and practical computer use alongside clearer natural language generation. Positioned as an upgrade to GPT‑5.6 Terra, the new model targets everyday workplace tasks such as drafting proposals, interpreting technical drawings, filling forms, and editing video.
Performance and cost tradeoffs
Terra continues OpenAI’s emphasis on cost‑intelligence across model families: standard API pricing places GPT‑6.1 Terra 50% lower than GPT‑5.6 Terra and GPT‑6.1 Sol, making it suitable for high‑volume automation like reading incoming invoices and entering them into accounting software. In ChatGPT Work and Codex, customer plan allowances stretch further when using GPT‑6.1 Terra compared with GPT‑5.6 Terra and substantially more than with GPT‑6.1 Sol or GPT‑6 Astra.
Announced standard API pricing:
- $1 per million input tokens
- $0.05 per million cached input tokens
- $5 per million output tokens
Benchmarks — computer use and visual reasoning
Agents’ Last Exam (professional multi‑step tasks):
- At medium reasoning effort, GPT‑6.1 Terra scores 15 percentage points higher than GPT‑5.6 Terra while costing roughly one‑third as much.
- In some measures Terra performs on par with GPT‑6 Astra at about one‑tenth the cost per token.
- At maximum effort, Terra is over three times as token‑efficient as Claude Opus 5.5 and about one‑quarter the price per token, yielding roughly a 9× lower cost per task.
- Terra is also faster: it completes tasks more than three times quicker than both GPT‑6 Astra and GPT‑6.1 Sol.
MMMU‑Pro (visual reasoning):
- GPT‑6.1 Terra outperforms GPT‑5.6 Terra at every reasoning setting.
- It approaches the scores of GPT‑6 Astra and GPT‑6.1 Sol while operating at much lower cost.
OSWorld 2.1 (desktop app usage for multi‑step tasks):
- At medium effort, GPT‑6.1 Terra exceeds the best score of GPT‑5.6 Terra at a fraction of the per‑task cost.
Reduced hallucinations and error rates
OpenAI reports focused investment this year on decreasing hallucinations and factual errors. On an internal benchmark of error‑inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT‑5.6 Terra to 16.7% with GPT‑6.1 Terra — a nearly 37 percentage‑point reduction. At maximum effort, the error rate dropped by 18 percentage points to 4.3%, nearly matching the rates for GPT‑6 Astra and GPT‑6.1 Sol.
Software engineering and scientific coding
Terminal‑Bench 4.0 (command‑line coding tasks):
- At maximum reasoning effort GPT‑6.1 Terra solves 50.3% of tasks, more than double the published 21.5% score of GPT‑5.6 Terra.
- For the most complex engineering problems, GPT‑6.1 Sol and GPT‑6 Astra remain stronger options.
Terminal‑Bench‑Science 0.1 (code‑based scientific research tasks):
- At maximum effort GPT‑6.1 Terra outperforms GPT‑6.1 Sol at lower cost, and comes within five points of Claude Opus 5.5 while costing less than one‑fifth of the estimated expense.
Safety, alignment, and availability
OpenAI says it continues to train, test, and refine safeguards. GPT‑6.1 Terra meets the Critical threshold in cybersecurity under OpenAI’s Preparedness Framework and retains the protections developed for GPT‑6 Astra. The model incorporates the GPT‑6 series’ alignment improvements: in OpenAI’s evaluations, GPT‑6.1 Terra is more likely to remain on‑task, disclose limitations honestly, and resist prompt‑injection attacks compared to GPT‑5.6 Terra.
Availability
Starting today, GPT‑6.1 Terra is available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can access it via the OpenAI API under the name gpt‑6.1‑terra.
Summary
GPT‑6.1 Terra is presented as a balanced, cost‑efficient option for automating daily work that requires visual reasoning and tool use. According to OpenAI’s published benchmark results, the model delivers notable improvements in accuracy, speed, and token efficiency relative to GPT‑5.6 Terra while maintaining the safety and alignment features of the GPT‑6 family.
