AI Model Finder

Find the right LLM for the job.

AI Model Finder scores 121 models on independent benchmarks, rates how well each fits your use case, and works out whether an open model will run on your hardware.

Top models overall

  1. GPT-6 Astra · OpenAI · overall 93/100
  2. Claude Opus 5.5 · Anthropic · overall 92/100
  3. Claude Sonnet 5.5 · Anthropic · overall 86/100
  4. Claude Fable 5.1 · Anthropic · overall 85/100
  5. Gemini 4 Argon · Google · overall 83/100
  6. GPT-6.1 Sol · OpenAI · overall 83/100
  7. Claude Opus 5 · Anthropic · overall 81/100
  8. GPT-6.1 Terra · OpenAI · overall 77/100
  9. Claude Opus 4.5 · Anthropic · overall 77/100
  10. GPT-5 · OpenAI · overall 76/100
  11. Qwen3.8-Flash-Next · Alibaba (Qwen) · overall 75/100
  12. Qwen3.8-Max · Alibaba (Qwen) · overall 75/100
  13. Kimi K3 · Moonshot AI · overall 73/100
  14. Kimi K2.6 · Moonshot AI · overall 72/100
  15. o3 · OpenAI · overall 71/100

Top open-weight models

  1. Qwen3.8-Flash-Next · Alibaba (Qwen) · overall 75/100
  2. Kimi K3 · Moonshot AI · overall 73/100
  3. Kimi K2.6 · Moonshot AI · overall 72/100
  4. GLM-5.3 · Z.ai (Zhipu) · overall 70/100
  5. GLM-5.3-Flash · Z.ai (Zhipu) · overall 67/100
  6. DeepSeek-V3.1 · DeepSeek · overall 66/100
  7. DeepSeek-V4.1-Flash · DeepSeek · overall 65/100
  8. MiMo-V2.6-Pro · Xiaomi · overall 65/100
  9. Qwen3.8-2.4T-A95B · Alibaba (Qwen) · overall 64/100
  10. Nemotron 3 Ultra · NVIDIA · overall 63/100

Best models by use case

General assistant

Everyday Q&A, explanations, summaries and broad knowledge.

  1. GPT-6 Astra · OpenAI · fit 97/100
  2. Claude Opus 5.5 · Anthropic · fit 94/100
  3. Claude Sonnet 5.5 · Anthropic · fit 88/100
  4. GPT-6.1 Sol · OpenAI · fit 88/100
  5. GPT-5 · OpenAI · fit 84/100

Coding

Writing, fixing and reviewing code, from snippets to whole repositories.

  1. Claude Opus 5.5 · Anthropic · fit 89/100
  2. Claude Sonnet 5.5 · Anthropic · fit 89/100
  3. Gemini 4 Argon · Google · fit 88/100
  4. Claude Opus 4.5 · Anthropic · fit 86/100
  5. GPT-6 Astra · OpenAI · fit 86/100

Math & reasoning

Multi-step math, logic and science problem solving.

  1. GPT-6 Astra · OpenAI · fit 95/100
  2. Claude Fable 5.1 · Anthropic · fit 94/100
  3. Claude Opus 5.5 · Anthropic · fit 92/100
  4. Claude Opus 5 · Anthropic · fit 86/100
  5. Kimi K3 · Moonshot AI · fit 86/100

Agents & tool use

Autonomous multi-step tasks, function calling and computer/terminal use.

  1. Claude Opus 4.5 · Anthropic · fit 90/100
  2. Claude Sonnet 5.5 · Anthropic · fit 89/100
  3. GPT-6 Astra · OpenAI · fit 89/100
  4. Claude Opus 5.5 · Anthropic · fit 86/100
  5. Gemini 4 Argon · Google · fit 85/100

Expert research

PhD-level questions, scientific analysis and frontier knowledge work.

  1. Claude Fable 5.1 · Anthropic · fit 100/100
  2. Claude Opus 5.5 · Anthropic · fit 98/100
  3. GPT-6 Astra · OpenAI · fit 97/100
  4. Claude Opus 5 · Anthropic · fit 96/100
  5. Claude Sonnet 5.5 · Anthropic · fit 87/100

Vision & documents

Understanding images, charts, screenshots and scanned documents.

  1. Claude Opus 5.5 · Anthropic · fit 97/100
  2. GPT-6 Astra · OpenAI · fit 96/100
  3. Gemini 3.8 Flash · Google · fit 95/100
  4. GPT-5 · OpenAI · fit 95/100
  5. Kimi K3 · Moonshot AI · fit 95/100

Long documents & RAG

Working over large documents, codebases or retrieved context. Scored on context window size and AA-LCR long-context reasoning.

  1. Claude Fable 5.1 · Anthropic · fit 100/100
  2. Claude Opus 5 · Anthropic · fit 100/100
  3. Claude Opus 5.5 · Anthropic · fit 100/100
  4. Claude Sonnet 5.5 · Anthropic · fit 100/100
  5. DeepSeek-V4.1-Flash · DeepSeek · fit 100/100

Writing & content

Drafting, rewriting and style-following. No dedicated benchmark, so this uses instruction following, overall intelligence and curated tags.

  1. Claude Opus 5.5 · Anthropic · fit 93/100
  2. GPT-6 Astra · OpenAI · fit 91/100
  3. Claude Sonnet 5.5 · Anthropic · fit 87/100
  4. GPT-5 · OpenAI · fit 84/100
  5. Claude Opus 5 · Anthropic · fit 83/100

Multilingual

Working across many languages. Uses curated tags plus general knowledge.

  1. Claude Opus 5.5 · Anthropic · fit 92/100
  2. GPT-6 Astra · OpenAI · fit 90/100
  3. DeepSeek-V3.1 · DeepSeek · fit 87/100
  4. GPT-4o (Aug 2024) · OpenAI · fit 87/100
  5. GPT-6.1 Sol · OpenAI · fit 86/100

On-device / edge

Running locally on phones, laptops or small GPUs for privacy and offline use. Scored mostly on memory footprint.

  1. Qwen3.5-4B · Alibaba (Qwen) · fit 88/100
  2. Llama 3.2 1B · Meta · fit 84/100
  3. Llama 3.2 3B · Meta · fit 83/100
  4. Gemma 3 1B · Google · fit 78/100
  5. Qwen3.5-0.8B · Alibaba (Qwen) · fit 78/100

Cybersecurity

Defensive security work: secure code review, threat and log analysis, incident response and CTF practice. Uses security benchmarks where a score is reported, and hands-on terminal and coding results otherwise.

  1. Claude Opus 5.5 · Anthropic · fit 81/100
  2. Claude Sonnet 5.5 · Anthropic · fit 81/100
  3. GPT-6 Astra · OpenAI · fit 80/100
  4. Claude Opus 4.5 · Anthropic · fit 79/100
  5. GPT-6.1 Sol · OpenAI · fit 76/100

Overall is the average fit for general use, coding and reasoning, from benchmark scores. How models are scored.