LLM Models · Open Source
Browse the world's large language models. Compare parameters, benchmarks, VRAM and more.
Open-source model listings are rows marked as open weights or open code. The mark is a filter, not a grant of rights. The license field and the upstream license file decide whether you may serve the weights, fine-tune them, or ship them in a product. If the license field is empty, the page does not know, and you should not guess from the badge.
Choose this category when the prompt contains source or customer text that must stay on a network you control, when you need a build you can hash, or when the hosted bill would exceed a machine you already run. The cost is operations: memory, a runtime, and a person who notices the process died. VRAM, when the page has it, is the first sizing hint. Parameter count is only a size class. Mixture-of-experts rows can quote a huge total and activate less. How to read a model listing walks those fields in order.
Do not assume a brand is open because one row with that brand is open. DeepSeek and Qwen appear as weights and as hosted endpoints. The flag on the specific row is the fact. Open source and proprietary models is the decision guide, including the five lines worth handing to a teammate.
Alias rows named Latest are usually not this category. They are pointers a vendor can move. Pin a weight you downloaded.
A high directory score inside this filter still is not a benchmark we ran. Sort helps you open a page. The license file helps you decide. If you cannot operate a GPU this month, a closed listing may be the honest choice. Pretending an open model is "installed" because you read the card is how a design doc lies.
Store the weight hash next to the license note. A later download with a different hash is a different model, even when the card title did not change. Re-run the one prompt you graded before you let the new file replace the old one.
📰 Tech Blog 📄 Full Report
Kimi K3.1 Pro is Moonshot AI's flagship open-weight multimodal reasoning model and a post-trained successor to Kimi K3, with a 1M-token context window. It is built for long-horizon coding, 3D...
👋 Join our WeChat or Discord community. 📖 Check out the GLM-5.2 blog and GLM-5 Technical report . 📍 Use GLM-5.2 API services on Z.ai API Platform. 🔜 Try GLM-5.2 here .
👋 Join our WeChat or Discord community. 📖 Check out the GLM-5.2 blog and GLM-5 Technical report . 📍 Use GLM-5.2 API services on Z.ai API Platform. 🔜 Try GLM-5.2 here .
Description: The NVIDIA GLM-5.2 NVFP4 model is the quantized version of ZAI’s GLM-5.2 model, which is an auto-regressive language model that uses an optimized transformer architecture. GLM-5.2 is a Mixture-of-Experts (MoE) model for reasoning and coding that uses sparse attention…
Description: The NVIDIA DeepSeek-V4-Pro-NVFP4 model is the quantized version of the DeepSeek-V4-Pro model, which is a Mixture-of-Experts (MoE) language model with 1.6 trillion total parameters and 49 billion activated parameters. For more information, please check here. The NVIDI…
DeepSeek-V4-Flash-0731 is the official release of DeepSeek-V4-Flash , superseding the preview version, with substantially enhanced agentic capabilities. It has the same model structure as DeepSeek-V4-Flash-DSpark, i.e. it comes with a speculative decoding module attached.
Description: The NVIDIA DeepSeek-V4-Flash-NVFP4 model is a quantized version of DeepSeek AI's DeepSeek-V4-Flash model, an autoregressive Mixture-of-Experts language model that uses an optimized Transformer architecture with hybrid attention (Compressed Sparse Attention and Heavil…
DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
🤗 HuggingFace 📰 Blog 🎨 Xiaomi MiMo API Platform 🗨️ Xiaomi MiMo Studio 💻 Xiaomi MiMo Desktop
🤗 HuggingFace 📰 Blog 🎨 Xiaomi MiMo API Platform 🗨Xiaomi MiMo Studio 💻 Xiaomi MiMo Desktop
MiMo-V2.6-Pro is the flagship foundation model developed by Xiaomi. Built at a scale of over 1T parameters, it is designed to push the ceiling of capability for the most demanding...