Skip to content
TheChatArchiveThe Chat Archive
Perplexitysonar-reasoning· PerplexityHugging Face LMArena Search Arena 24K

what is the top ten model in livebench now?

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with gpt-4o-search-preview-high’s answer.
  1. User

    what is the top ten model in livebench now?

  2. Perplexity Assistant

    Top LLMs in LiveBench (March 2025)

    Based on recent benchmarks and performance metrics, here are the leading models in LiveBench as of March 2025:

    1. Claude 3.7 Sonnet (Anthropic)

      • Global Average Score: 76.10 (highest overall)[8].
      • Strengths: Hybrid reasoning, data analysis, and inference tasks[8].
      • Weaknesses: Slightly trails in pure reasoning and coding compared to OpenAI’s o3-mini-high[8].
    2. OpenAI o3-mini-high

      • Coding Performance: 82.74 (highest in coding benchmarks)[12].
      • Use Case: Advanced coding, logic, and structured tasks[12].
    3. OpenAI o1-preview

      • Global Average: ~75.88 (previous leader before Claude 3.7)[8].
      • Strengths: General-purpose tasks and reasoning[2][8].
    4. DeepSeek-R1

      • Open-Source Leader: Top open-source model in LiveBench’s top ten[10].
      • Strengths: Complex data analysis and reasoning[9].
    5. QwQ-32B (Tongyi Qianwen)

      • Open-Source Performance: Second Chinese model in the top ten[10].
      • Strengths: Mathematical reasoning and coding efficiency[7].
    6. Gemini 2.0 Pro

      • Coding & Context: Strong in large-context tasks and coding[5].
      • Use Case: Google Cloud integration and multi-file refactoring[5].
    7. GPT-4o

      • Workhorse: Fast, cost-efficient, and widely used in tools like GitHub Copilot[5].
      • Coding Accuracy: ~90% on HumanEval[5].
    8. Mistral Large 2

      • Multilingual & Coding: Outperforms Llama 3.1 405B in coding tasks[1][7].
      • Context Window: 128K tokens[1].
    9. Grok-3 (xAI)

      • Real-Time Reasoning: Excels in dynamic data processing and extended context (1M tokens)[1][9].
      • Modes: “Think” and “Big Brain” for problem-solving[1].
    10. Claude 3.5 Sonnet

      • Legacy Performance: Still competitive in coding and reasoning[2][5].

    Key Trends

    • Claude 3.7 Sonnet dominates overall performance with hybrid reasoning[8].
    • OpenAI’s o3-mini-high leads in coding-specific benchmarks[12].
    • Chinese Open-Source Models (QwQ-32B, DeepSeek-R1) rise in global rankings[10].

    For specialized tasks, Claude 3.7 Sonnet (general-purpose), o3-mini-high (coding), and DeepSeek-R1 (open-source) are top choices[5][8][10].

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