Skip to content
TheChatArchiveThe Chat Archive
o3-2025-04-16· OpenAIHugging Face LMArena Human Preference 140K

list top models by context window size

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Qwenqwen3-30b-a3b’s answer.
  1. User

    list top models by context window size

  2. Assistant

    🎯 Largest-context LLMs (publicly announced, May 2024)

    RankModelMax context window (tokens*)≈ WordsProvider / AvailabilityKey notes
    1Gemini 1.5 Pro1 000 000~750 000Google DeepMind (private preview API)“Long-context” architecture; can ingest full movies or large code bases
    2Claude 3 family (Opus / Sonnet / Haiku)200 000~150 000Anthropic (API)Same 200 k limit across tiers; streaming & tool-use ready
    3Claude 2.1200 000~150 000Anthropic (API)First mainstream 200 k release (Nov 2023)
    4Yi-34B-200K (open-source)200 000~150 00001.AI (HF download)Rotary-scaled; instruction-tuned checkpoints available
    5GPT-4 Turbo128 000~95 000OpenAI (API, ChatGPT “Turbo”)Price-tiered vs 8 k/32 k GPT-4
    6Cohere Command-R / R⁺128 000~95 000Cohere (API)Retrieval-augmented by design
    7GPT-4 (32k variant)32 000~24 000OpenAI (legacy API tier)Original GPT-4 launch model
    8Mistral-Large32 000~24 000Mistral (API)Flash-attention 2; 27B params
    9Mixtral 8×7B Instruct (long-rope)32 000~24 000Mistral (open-source)Community “long-rope” checkpoints reach 65 k+
    10Llama-2-70B-Long32 000~24 000Meta / community re-ropeOpen-source; slower inference
    11GPT-3.5-Turbo-16k16 000~12 000OpenAI (API)Budget option with extended window
    12Standard GPT-3.5-Turbo4 096~3 000OpenAI (API, free ChatGPT tier)Baseline for many apps

    *Token ≈ 0.75 word in English; 100 k tokens ≈ 75 k words or ~300 book pages.

    Quick takeaways • If you truly need huge documents (>128 k tokens) today, Gemini 1.5 and Claude 3 are the only turnkey choices; Yi-34B-200K is an open-source DIY option.
    • For most enterprise / RAG workloads, 32 k–128 k windows (GPT-4 Turbo, Cohere Command-R, Mistral-Large) already cover entire legal contracts, codebases, or multi-chapter manuals.
    • Very long windows do not guarantee perfect recall—chunking + retrieval still matters! 🧠🔍

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