Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-pro-exp-03-25-grounding· GoogleHugging Face LMArena Search Arena 24K

What model is the 24karatgold model on lmarena.ai

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Perplexitysonar-pro’s answer.
  1. User

    What model is the 24_karat_gold model on lmarena.ai

  2. Gemini Assistant

    Based on the search results and the nature of the LMSYS Chatbot Arena, "24_karat_gold" is an anonymous, experimental model being evaluated on the platform.

    Here's what we know:

    1. Anonymous Model: LMSYS often tests unreleased or experimental models under anonymous codenames on the Chatbot Arena. This allows them to gather real-world user preference data without revealing the model's identity or affiliation initially. "24_karat_gold" falls into this category, alongside other codenames like "Spider", "stradale", "roma", and "venom" mentioned in discussions[1].
    2. LMSYS Platform: The model is hosted and evaluated on lmarena.ai (Chatbot Arena), a platform run by LMSYS and UC Berkeley SkyLab to rank large language models based on human feedback.
    3. Speculation vs. Fact: While there might be speculation online about its origins (one source mentions a possible, though speculative, link to Meta's GPU clusters due to the name, but this is not confirmed), the actual underlying model architecture or provider for these anonymous test models is typically not disclosed by LMSYS during the evaluation phase.
    4. Purpose: The goal is to test the model's performance and gather user preference data before a potential public release or publication of results.

    Therefore, the specific model identity behind "24_karat_gold" is intentionally kept confidential by LMSYS as part of their anonymous testing protocol on the Chatbot Arena. It is an experimental model undergoing evaluation.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.

Identical first prompt