Google DeepMind has introduced a new model variant, Gemini 4 Carbon. The model has secured the top position on the LM Arena Text Arena leaderboard.
Gemini 4 Carbon achieved a score of 1,528 points in the Text Arena, which relies on a blind-voting system to evaluate human preferences for model responses.
Beyond text generation, the model demonstrated performance in specialized tasks, ranking second in the Agent Arena and fourth in the WebDev Arena.
The LM Arena platform utilizes an Elo-style rating system based on millions of head-to-head human votes to assess the performance of frontier AI models.
