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-mini-search-preview’s answer.
  1. User

    what is the top ten model in livebench now?

  2. Perplexity Assistant

    Top Ten Models in LiveBench (March 2025)

    1. Claude 3.7 Sonnet (Anthropic)

    • Global Average Score: 76.10 (highest overall)[8].
    • Strengths: Hybrid reasoning, data analysis (74.05), and inference tasks (81.25)[8].
    • Weaknesses: Slightly trails OpenAI in pure reasoning (87.83 vs. OpenAI’s 89.58)[8].

    2. OpenAI o3-mini-high

    • Coding Performance: 82.74 (highest in coding benchmarks)[10].
    • Strengths: Optimized for STEM reasoning and logic, outperforms o1 and Claude 3.5 Sonnet in coding[10].
    • Availability: Free access via ChatGPT’s Reason button; Pro users get unlimited access[10].

    3. OpenAI o1-preview

    • Global Average: ~75.88 (slightly behind Claude 3.7 Sonnet)[8].
    • Strengths: Strong in reasoning (91.58) and language tasks[2][8].

    4. DeepSeek R1

    • Coding Accuracy: ~49% on SWE-bench (comparable to OpenAI’s o1)[5].
    • Strengths: Blends coding and reasoning with MoE architecture[5].

    5. GPT-4o (OpenAI)

    • Coding Accuracy: ~90% on HumanEval[5].
    • Strengths: Fastest high-accuracy model for real-time coding assistance[5].

    6. Claude 3.5 Sonnet (Anthropic)

    • Coding Accuracy: 92% on HumanEval[5].
    • Strengths: Low hallucination rate and strong reasoning[5].

    7. Grok-3 (xAI)

    • Context Window: 1 million tokens[1].
    • Strengths: Real-time data processing and advanced reasoning modes (Think/Big Brain)[1][9].

    8. Gemini 2.0 Pro (Google)

    • Strengths: Large-context tasks (up to 32K tokens) and integration with Google Cloud[5][9].

    9. Mistral Large 2

    • Coding Performance: Outperforms Llama 3.1 405B in multiple languages[1].
    • Strengths: Multilingual capabilities and reduced hallucinations[1].

    10. DeepSeek Chat V3

    • Coding Accuracy: ~90% on HumanEval (estimated)[5].
    • Strengths: Efficient for structured problem-solving[5].

    Key Trends

    • Claude 3.7 Sonnet dominates overall performance with hybrid reasoning[8].
    • o3-mini-high leads in coding-specific benchmarks[10].
    • Specialization vs. Versatility: Models like Claude 3.7 Sonnet excel in balanced performance, while o3-mini-high and GPT-4o target niche areas like coding[8][10].

    For coding, o3-mini-high and Claude 3.7 Sonnet are top choices, while Claude 3.7 Sonnet remains the most versatile general-purpose model[5][8][10].

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