Source: OpenAI API release documentation and OpenRouter model registry metadata published October 26, 2026.
Engineered to serve as a fast subagent workhorse, GPT-6.1 Terra targets latency-sensitive workflows that involve parsing complex graphical user interfaces, technical schematics, and tabular reports. In contrast to flagship models like GPT-6.1 Sol that employ deep chain-of-thought exploration across every token, Terra uses a calibrated inference architecture that accelerates token output by 3x on structured tasks such as form completion, browser automation, and document categorization.
The model's native context window spans 1,050,000 tokens, enabling developers to retain full execution traces, repository snapshots, and visual state histories within a single ongoing context session. To support continuous agentic loops, prompt cache reads are billed at $0.05 per million tokens—a 95% reduction from base input rates. Standard input is priced at $1.00 per million tokens, while completion generation costs $5.00 per million tokens.
On computer-use and visual benchmarks, Terra posts competitive figures against heavier systems. Evaluation on OSWorld and MMMU-Pro demonstrates that the model reliably translates visual UI elements into coordinate clicks, keyboard shortcuts, and form fills without suffering hallucinated layout coordinates. The model also introduces dedicated handling for high-resolution document feeds, preserving OCR accuracy across nested tables and dense diagrams without downsampling artifacts.
OpenAI is making Terra the default driver for routine background workers in ChatGPT Work workspaces and enterprise Copilot integrations. For batch asynchronous processing—such as overnight code reviews, database hygiene, and catalog enrichment—the batch endpoint prices input at $0.50 per million tokens and output at $2.50 per million tokens, undercutting comparable mid-tier offerings from Anthropic and Google.
While Terra excels at rapid perception and deterministic function calling, developers noted that it intentionally truncates exploratory mathematical reasoning compared to Sol-level models. Complex algorithmic synthesis and ambiguous system architecture tasks remain better served by frontier reasoning tiers, cementing Terra's role as a pragmatic operational engine.