Model guide · updated 2026

Best open-source AI models

Open-weight models have closed much of the gap with closed flagships — at a fraction of the price. These are the strongest open models by intelligence, from DeepSeek, Zhipu (GLM), Alibaba (Qwen), Moonshot (Kimi), MiniMax and OpenAI's GPT-OSS.

  1. 1

    DeepSeek v3.1

    DeepSeekTop pick

    Software engineering

    AA 60
    Intelligence
  2. 2

    Kimi K2.6

    Moonshot AI

    Long-horizon autonomous coding and multi-file refactoring projects

    AA 54
    Intelligence
  3. 3

    DeepSeek V4 Pro

    DeepSeek

    High-volume automated coding pipelines

    AA 52
    Intelligence
  4. 4

    GLM 5.1

    Zhipu AI

    Autonomous software engineering agents that run multi-step tasks for hours without human checkpoints

    AA 51
    Intelligence
  5. 5

    Qwen 3.6 Plus

    Alibaba

    Agentic coding and repository-level problem solving

    AA 50
    Intelligence
  6. 6

    MiniMax M2.7

    MiniMax

    Agentic coding workflows and multi-turn tool-calling chains where the model drives autonomous

    AA 50
    Intelligence
  7. 7

    GLM 5

    Zhipu AI

    Code generation and completion

    AA 50
    Intelligence
  8. 8

    DeepSeek V4 Flash

    DeepSeek

    High-volume coding assistants and IDE integrations where LiveCodeBench 91.6% and SWE-bench 79% performance matters at low per-token cost

    AA 47
    Intelligence

Open-weight models only (GPT-OSS, DeepSeek, GLM, Qwen, Kimi, MiniMax), ranked by the Artificial Analysis intelligence index.

Frequently asked questions

Here we mean open-weight models — ones whose trained weights are publicly released so they can be self-hosted or run through many providers. DeepSeek, GLM, Qwen, Kimi, MiniMax and OpenAI's GPT-OSS all qualify.

The top open models now rival closed flagships on many benchmarks, especially coding and math, while costing far less. They can still trail the very best closed models on the hardest reasoning tasks.

Yes — on just4o.chat the open models above run through hosted inference, so you get the cost benefits without managing GPUs.