OpenAI is actively seeding new models on LM Arena, a platform where users can benchmark and compare top-tier models before they are widely released. The latest arrival, Robin High, stands out as OpenAI’s most capable candidate in the current lineup. This model successfully solves challenging math tasks that have previously been the benchmark for Gemini 3 Pro and GPT-5.1 Pro. In current tests, only Gemini 3 Pro and Robin High can consistently handle these problems without additional computational resources, pointing to Robin High as OpenAI’s response to Google’s Gemini 3 Pro and possibly signalling its position as a next-generation flagship.
There are also references to another internal project, Garlic, which has surfaced in both internal leaks and media speculation as a possible code name for OpenAI’s next major release.
— ChatGPT (@ChatGPTapp) December 10, 2025
Although the direct link between Robin High and Garlic is not yet confirmed, recent activity and hints from OpenAI suggest a coordinated rollout of these advanced models. For researchers and developers following the model landscape, Robin High is now accessible for evaluation on LM Arena, offering a useful preview of OpenAI’s direction in competing with Google’s latest.
These additions reflect OpenAI’s strategy to maintain momentum in model development and keep pace with rapid advancements from competitors.
