Command Palette

Search for a command to run...

Best Open LLMs (2025)

Stephen M. Walker II

A round-up of widely praised open-source language models launched in 2024 and 2025 with key release links.

Open-source models keep advancing quickly. Below are widely discussed releases from 2024 and 2025 with links to learn more.

ModelHighlights
Mistral Small 3.224B latency-optimised multilingual model with stronger instruction following and Apache-2.0 weights.1
Llama 3Meta's permissive-licensed model family with strong general benchmarks.23
Mixtral 8x22BMistral's sparse MoE that trades parameters for speed while retaining accuracy.4
Gemma 3Google's lightweight multimodal family (1–27B) with 128K context and broad language coverage.5
Qwen3Alibaba's next-generation family with improved scaling and open weights.6
DeepSeek-R1Reinforcement-learned 37B-active-parameter reasoning model with 128 K context and MIT license.7
OLMo 2Allen AI's fully reproducible research model with transparent training pipeline.8
InternLM2-Chat-20BLong-context (128 K) chat model that excels at reasoning tasks.9
Yi-1.534B multilingual model that also performs well on code tasks.10
Phi-3-mini4.3B small-footprint model that punches above its size.11
GLM-4-9B-ChatBilingual English/Chinese 9B model with competitive open benchmarks.12

  References

  References

  1. huggingface.co ↩

  2. ai.meta.com ↩

  3. huggingface.co ↩

  4. mistral.ai ↩

  5. blog.google ↩

  6. github.com ↩

  7. huggingface.co ↩

  8. allenai.org ↩

  9. huggingface.co ↩

  10. huggingface.co ↩

  11. huggingface.co ↩

  12. huggingface.co ↩