Show HN: An opinionated ranking of 21 open-weight LLMs, filterable by your GPU
A ranking of 21 open-weight large language models has been published, allowing users to filter results based on their specific GPU hardware to find the best-performing model for their setup.
Background
- The article ranks 21 open-weight large language models (LLMs) — AI models whose internal parameters ("weights") are publicly released, so anyone can download and run them locally rather than relying on a paid cloud service like ChatGPT.
- "Open-weight" is not the same as fully open-source; the training data and code are often kept secret, but the model itself can be self-hosted, which matters for privacy, cost control, and censorship avoidance.
- The ranking is "opinionated" and "filterable by your GPU" — meaning it helps a user pick a model that will actually run on their specific graphics card (VRAM is the key constraint). Consumer GPUs (e.g., NVIDIA RTX 3090 with 24GB) can't run the biggest models, which need hundreds of GB of server-grade memory.
- Key entrants: Meta's Llama 3, Mistral's Mixtral, Alibaba's Qwen, Google's Gemma, and Microsoft's Phi — each with different strengths, sizes, and permissive licenses. Some ban commercial use or require attribution.
- The post appeared on Hacker News ("Show HN"), a part of Y Combinator's forum where makers show projects they built. This signals the author is an independent developer or small team, not a big corporation.