Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-opus-4-20250514-thinking-16k· AnthropicHugging Face LMArena Human Preference 140K

can you always see the model name after you choose which answer is better

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.5-pro-preview-05-06’s answer.
  1. User

    can you always see the model name after you choose which answer is better

  2. Claude Assistant

    No, you don't always see the model names after voting in LM Arena. Here's how it typically works:

    Standard Behavior:

    • During comparison: Models are shown as "Model A" and "Model B" - completely anonymous
    • After voting: In most cases, yes, the model identities are revealed immediately after you submit your vote (e.g., you'll see that Model A was "GPT-4-Turbo" and Model B was "Claude 3 Opus")

    Exceptions Where Names Might Not Be Revealed:

    1. Private/Experimental Models: If a model developer has requested anonymity for their unreleased model, it might show up as something generic like "anonymous_model_1" or "experimental_v3" even after voting

    2. Arena Modes: LM Arena has experimented with different modes over time. Some modes might:

      • Delay the reveal until after multiple rounds
      • Keep certain models anonymous longer for research purposes
      • Have special testing configurations
    3. Technical Issues: Occasionally, there might be glitches where the reveal doesn't work properly

    4. Special Research Studies: Sometimes LMSys might run special experiments where they intentionally don't reveal model names to study voting patterns

    Why This Matters:

    • The immediate reveal helps users learn which models they preferred in practice
    • It provides transparency about what's being tested
    • For private models, continued anonymity protects competitive information while still allowing benchmarking

    The platform's design philosophy generally favors transparency after voting, but there are legitimate cases where full model names aren't disclosed to protect proprietary information or for research purposes.

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