OpenAI announced GPT-6.1 Terra, a model optimized for computer interaction, visual reasoning, coding, and professional tasks, priced at about half the cost of GPT-5.6 Terra via API.

Terra shows strong performance, matching or approaching GPT-6.1 Sol and GPT-6 Astra on multiple benchmarks, including 85.5% on MMMU-Pro Vision (max effort) with results varying on other tests like Terminal-Bench 4.0 and OSWorld offline sets.

Under higher reasoning settings, Terra reportedly delivers substantially lower per-token costs and faster task completion than GPT-6 Astra and GPT-6.1 Sol, with claims of up to ninefold cost reductions for some tasks.

Readers are invited to suggest new channels in the comments.

OPEN ROUTER integration and a rolling 24-hour data window power live listings of new models and comparisons on MODEL FACE-OFF channel 13.

DEV∙TV offers a retro TV interface for developers with 14 channels, designed as a single HTML file with no infrastructure and channel-flipping like tabs.

Independent verification of OpenAI’s claims isn’t yet available; third-party tests are needed, and potential admin controls on Business/Enterprise plans are noted.

No market data on token price or a specific future catalyst date was provided for the Terra launch.

Terra’s use cases span transferring invoices to accounting software, autofilling forms, drafting proposals, handling technical drawings, converting architectural plans to 3D models, and creating time-stamped logs for animations.

Security features reach the Critical level in OpenAI’s Preparedness Framework, aiming to guard against prompt-injection and keep tasks within scope, though real-world effectiveness depends on deployment.

All scores are OpenAI-reported; prices and hardware are undisclosed, with reference links to the announcement, API model page, and system card addendum.

Terra supports Streaming, Function calling, and Structured outputs, but does not support Fine-tuning.