OpenAI Launches GPT-6.1 Terra for Public Testing on Arena.ai Platforms

OpenAI has rolled out the GPT-6.1 Terra model for public testing on its Agent Arena and Code Arena platforms, which are operated by Arena.ai. This initiative enables users to assess the model's capabilities in handling both agentic tasks and programming work.

Previously, GPT-6.1 Terra was referenced internally by Microsoft as a judge model for coding evaluations across 18 configurations. The introduction of Terra is particularly intriguing given its prior role in evaluating performance, as it now faces the challenge of being judged itself.

Agent Arena, which tracks over 2.5 million sessions, employs causal tracing methodologies to evaluate AI performance, while Code Arena focuses on programming tasks. The performance of Terra will be compared to its sibling model, GPT-6.1 Sol, which was launched for community evaluation on September 29, 2026. Sol has been noted for its efficiency, achieving performance metrics close to the top-tier GPT-6 Astra variant at a significantly lower cost.

The performance of Terra in these arenas will be closely monitored, particularly in terms of its ranking and cost per task compared to Sol's range of approximately $0.56 to $0.82. This testing phase may provide insights into the model's real-world applicability and its potential for broader availability.