排行榜

  1. https://artificialanalysis.ai/
  2. https://llm-stats.com/
  3. https://www.swebench.com/
  4. https://openrouter.ai/rankings
  5. https://pinchbench.com/
  6. https://kilo.ai/leaderboard
  7. https://llmdb.com/
  8. https://crfm.stanford.edu/helm/lite/latest/#/leaderboard

截止 2026-03 的领先模型 和价格:

序号模型参数规模输入价格($/1M)输出价格($/1M)窗口
1GPT-5.42.5151m
2Claude Opus 4.65251m
3Gemini 3.1 Pro2121m
4GLM-5744B A40B0.722.3200k
5MiMo-V2-Pro1000B A42B131m
6Minimax M2.7230B A10B0.31.2200k
7Kimi K2.51000B A32B0.452.2256k
8Qwen3.5397B A17B0.392.34256k
27B0.1951.56
122B A10B0.262.08
9Minimax M2.5230B A10B0.21.17200k
10DeepSeek V3.2685B A37B0.260.38128k
11Grok 4315
12MiMo-V2-Flash309B A15B0.090.29256k
13NVIDIA Nemontron 3 Super120B A12B BF16
14gpt-oss-120B0.0390.19128k
15gpt-oss-20B0.030.11128k
16Step-3.5-Flash196B A11B0.10.3256k
17LongCat-Flash-Thinking560B A27B0.090.29256k
18Mistral Large 3
19Google Gemma-4-31b31B0.140.40256k
20Cognition SWE-1.5
21NVIDIA Nemotron 3 Nano30B A3B
22K2 Think V2
23Olmo 32B32B

备注：NVIDIA Nemontron 和 K2 Think 开放程度高，Olmo 是示范用途，完全开放。更早的开源开放模型有 Pythia、LLM360、OpenCoder、BLOOM。

价格

  1. https://openrouter.ai/models
  2. https://artificialanalysis.ai/
  3. https://models.litellm.ai/
  4. https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json
  5. https://www.dmxapi.cn/rmb
  6. https://llm-stats.com/leaderboards/llm-leaderboard
  7. https://whatllm.org/
  8. https://pricepertoken.com/
  9. https://llm.minprices.com/
  10. https://github.com/syaoranwe/LLM-Price

运行

CLI

  1. https://github.com/ggerganov/llama.cpp
  2. https://github.com/foldl/chatllm.cpp
  3. https://github.com/ggerganov/whisper.cpp
  4. https://github.com/google/gemma.cpp
  5. https://github.com/EricLBuehler/mistral.rs
  6. https://github.com/ml-explore/mlx-lm

Desktop

  1. https://ollama.com
  2. https://lmstudio.ai
  3. https://cherry-ai.com
  4. https://msty.app
  5. https://github.com/open-webui/open-webui
  6. https://github.com/lobehub/lobe-chat
  7. https://github.com/chatboxai/chatbox
  8. https://github.com/danny-avila/LibreChat
  9. https://anythingllm.com
  10. https://www.nomic.ai/gpt4all
  11. https://github.com/lostruins/koboldcpp
  12. https://github.com/Mozilla-Ocho/llamafile

Server

  1. https://github.com/vllm-project/vllm
  2. https://github.com/sgl-project/sglang
  3. https://github.com/triton-inference-server/tensorrtllm_backend https://github.com/NVIDIA/TensorRT-LLM
  4. https://github.com/skypilot-org/skypilot
  5. https://github.com/huggingface/text-generation-inference
  6. https://github.com/lm-sys/FastChat
  7. https://github.com/bentoml/OpenLLM
  8. https://haystack.deepset.ai

Proxy

  1. https://github.com/BerriAI/litellm
  2. https://github.com/maximhq/bifrost
  3. https://github.com/tensorzero/tensorzero
  4. https://github.com/QuantumNous/new-api
  5. https://github.com/songquanpeng/one-api
  6. https://github.com/helicone/helicone
  7. https://portkey.ai/

SDK

  1. https://github.com/badlogic/pi-mono 21.1k stars
  2. https://github.com/pydantic/pydantic-ai 15.6k
  3. https://github.com/langchain-ai/deepagents 13.6k stars
  4. https://github.com/openai/openai-agents-python
  5. https://github.com/openai/openai-agents-js
  6. https://github.com/anthropics/claude-agent-sdk-python
  7. https://github.com/anthropics/claude-agent-sdk-typescript
  8. https://google.github.io/adk-docs/ 支持 Java, Go, TypeScript, Python
  9. https://ai-sdk.dev/ Vercel AI SDK for TypeScript
  10. https://www.llamaindex.ai/
  11. https://agentscope.io/
  12. https://github.com/microsoft/agent-framework

智能体

Github star 数量统计于 2026-03-08。

  1. https://github.com/openclaw/openclaw 278.6k stars
  2. https://github.com/ultraworkers/claw-code 158k stars
  3. https://github.com/HKUDS/nanobot 30.4k stars
  4. https://github.com/bytedance/deer-flow 25.9k stars
  5. https://github.com/zeroclaw-labs/zeroclaw 24.5k stars
  6. https://github.com/sipeed/picoclaw 22.9k stars
  7. https://github.com/badlogic/pi-mono 21.1k stars
  8. https://github.com/qwibitai/nanoclaw 20.3k stars
  9. https://github.com/AstrBotDevs/AstrBot 19.8k stars
  10. https://github.com/langbot-app/LangBot 15.5k stars
  11. https://github.com/RightNow-AI/openfang 12.4k stars
  12. https://github.com/agentscope-ai/CoPaw 9.6k stars
  13. https://github.com/nearai/ironclaw 7.3k stars
  14. https://github.com/nullclaw/nullclaw 5.9k stars
  15. https://github.com/HKUDS/OpenHarness 5.2k stars
  16. https://github.com/microclaw/microclaw 507 stars
  17. https://github.com/qhkm/zeptoclaw 433 stars
  18. https://github.com/loongclaw-ai/loongclaw 28 stars

应用

  1. 知识问答
    1. 智谱清言： https://chatglm.cn/ 、 https://z.ai ，支持 agent
    2. Minimax： https://agent.minimaxi.com/ ，只有 agent 模式，原来的 https://chat.minimaxi.com 域名已经跳转到 agent 域名
    3. Kimi: https://kimi.com ， agent 模式收费，每月三次免费额度
    4. 千问： https://qianwen.com ， https://qwen.ai ，有「深度研究」和「任务助理」，后者每月限制 20 次免费额度
    5. 腾讯元宝： https://yuanbao.tencent.com/
    6. DeepSeek： https://chat.deepseek.com/
    7. 小米 Mimo： https://aistudio.xiaomimimo.com/
    8. 字节豆包： https://www.doubao.com/ ， 有 「深入研究」和「超能模式」
    9. 美团龙猫： https://longcat.chat/ ， 有「深度研究」
    10. 阶越 AI： https://www.stepfun.com/
    11. ChatGPT： https://chatgpt.com/
    12. Nvidia： https://build.nvidia.com/openai/gpt-oss-120b
  2. 代码阅读：
    1. https://deepwiki.com/
    2. https://zread.ai/
  3. 个人桌面助理
    1. Anthropics Claude CoWork
      1. 基于 OpenCode 的 OpenWork
      2. 基于 OpenCode 的 Accomplish，曾用名 OpenWork
      3. Open Cowork
      4. Proma
      5. Craft Agents
      6. Kuse Cowork
      7. Eigent
      8. Open Claude Cowork
      9. OpenCowork
      10. Agent Cowork
    2. OpenClaw 包装或仿品
      1. 桌面版
        1. 腾讯 WorkBuddy(非 OpenClaw)，QClaw
        2. 阿里 QoderWork(非 OpenClaw)、悟空（非 OpenClaw)
        3. 字节 Trae Solo（非 OpenClaw）、UI TARS Desktop(非 OpenClaw)
        4. 智谱 AutoClaw
        5. 网易 LobsterAI
        6. ClawX
      2. 云版
        1. 阿里 JVSClaw
        2. 腾讯 ClawPro
        3. Moonshot KimiClaw
        4. Minimax MaxClaw
        5. 字节 ArkClaw, Aily
        6. 百度 DuClaw
    3. AionUi
    4. 阶跃 AI 桌面伙伴
    5. 昆仑天工 Skywork Desktop
  4. CLI
    1. open-interpreters
    2. opencli
    3. CLI-Anything

技能

  1. https://github.com/msitarzewski/agency-agents

训练

  1. Pre-training & fine-tuning
    1. https://github.com/hiyouga/LLaMAFactory
    2. https://github.com/modelscope/ms-swift
    3. https://github.com/deepspeedai/DeepSpeed
    4. https://github.com/NVIDIA/Megatron-LM
    5. https://github.com/hpcaitech/ColossalAI
    6. https://github.com/karpathy/nanochat
  2. Fine-tuning only
    1. https://github.com/unslothai/unsloth
    2. https://github.com/axolotl-ai-cloud/axolotl
    3. https://github.com/huggingface/trl
    4. https://github.com/lm-sys/FastChat
    5. https://github.com/verl-project/verl
    6. https://github.com/OpenRLHF/OpenRLHF
    7. https://github.com/meta-pytorch/torchtune
    8. https://github.com/inclusionAI/AReaL

参考

  1. https://en.wikipedia.org/wiki/Large_language_model
  2. https://github.com/Hannibal046/Awesome-LLM