Model Releases
OpenAI Releases GPT-6.1 Terra Multimodal Model
OpenAI's new GPT-6.1 Terra offers a massive leap in computer-use capabilities and visual reasoning at a 50% lower price point than its predecessors.
October 26, 2026
Model ReleaseGPT-6.1 Terra
Introduction
On October 26, 2026, OpenAI officially released GPT-6.1 Terra, a mid-sized multimodal model designed to bridge the gap between high-reasoning frontier models and high-speed agentic workflows. Positioned as the successor to GPT-5.6 Terra, this release marks a strategic shift toward making sophisticated computer-use and visual reasoning economically viable for large-scale automation.
For developers and AI engineers, Terra is not just another incremental update. It is a specialized tool optimized for the 'agentic era,' where models are expected to interact with operating systems, navigate complex UIs, and execute multi-step reasoning tasks with minimal latency. By combining improved natural language generation with precise visual grounding, OpenAI is targeting the core requirements of autonomous software agents.
- Successor to GPT-5.6 Terra
- Optimized for agentic workflows and computer-use
- Enhanced multimodal visual reasoning
- Significant reduction in token pricing
Key Features & Architecture
GPT-6.1 Terra is built on a refined multimodal architecture that prioritizes high-fidelity visual perception and low-latency execution. Unlike the larger GPT-6.1 Sol or Astra models, Terra is engineered for scale, allowing developers to deploy massive agentic swarms without the prohibitive costs typically associated with frontier-class reasoning.
The model demonstrates a significant leap in 'computer-use' capabilities. It can interpret screen pixels, understand hierarchical UI elements, and execute precise mouse and keyboard actions. This is coupled with a more natural, nuanced writing style that avoids the repetitive patterns often seen in earlier iterations of the GPT series. Furthermore, the model meets the Critical capability threshold in cybersecurity under OpenAI's Preparedness Framework, featuring robust resistance against prompt-injection attacks.
- Advanced computer-use integration for OS navigation
- High-speed task completion (3x faster than GPT-6 Astra/Sol)
- Enhanced prompt-injection resistance
- System card published for transparency and safety
Performance & Benchmarks
The benchmark data for GPT-6.1 Terra reveals a model that punches far above its weight class, particularly in technical and reasoning-heavy domains. In the MMMU-Pro visual reasoning benchmark, Terra outperforms GPT-5.6 Terra across every reasoning setting, nearly matching the performance of the much larger GPT-6 Astra and GPT-6.1 Sol models.