# The best free open-weight AI models Every option in the catalog for when you need to download weights you can inspect, fine-tune and self-host — 24 open-weight AI models, 24 of them free or freemium, with the pricing tier printed on every row. ## Stats - 24 tools listed - 24 free or freemium - 23 with no paid plan at all - 1 sub-categories covered ## FAQ ### What are the best free open-weight AI models? Flocci AI Tools lists 24 of them, ordered free tiers first: DeepSeek Models, ERNIE 4.5 Open Models (Baidu), Falcon (TII), Hunyuan (Tencent) and MiniMax M2 (open weights). Every one of them is free or has a free tier. Each entry states what it uniquely does rather than a score. ### Which of these are completely free? 23 on this page are marked fully free with no paid plan at all: DeepSeek Models, ERNIE 4.5 Open Models (Baidu), Falcon (TII), Hunyuan (Tencent) and MiniMax M2 (open weights). The remaining 1 are freemium, trial or paid, and each is labelled with its exact tier so nothing surprises you at sign-up. ### How is this list ordered? Free tools first, then freemium, then trial, then paid, and alphabetically inside each tier. Nobody pays to appear higher: Flocci AI Tools carries no ads, no sponsored slots and no paid listings, so the order reflects price and nothing else. ## Quick picks ### Which open-weight AI models work in a browser with no install? DeepSeek Models, ERNIE 4.5 Open Models (Baidu), Falcon (TII), Hunyuan (Tencent) and MiniMax M2 (open weights) all work in a browser tab with no install needed — just open the site and go. 12 of the 24 free open-weight AI models listed here are browser-based. ### Which open-weight AI models let you sign in with Google? Qwen (Alibaba) supports Google sign-in, so you can start without creating a new account. That's 1 of the 24 free open-weight AI models on this page. ### Which open-weight AI models have a mobile app? DeepSeek Models and Qwen (Alibaba) have an iOS or Android app. 2 of the 24 free open-weight AI models here can run on your phone. ### Which open-weight AI models can you self-host or run offline? GPT-OSS (OpenAI), Kimi K3 (Moonshot AI), Llama (Meta), Sarvam Open Models (30B / 105B) and TimesFM (Google Research) can be self-hosted or run locally without a browser — useful for privacy, air-gapped networks, or avoiding API rate limits. 5 of the 24 free open-weight AI models here run offline. ## Tools - [DeepSeek Models](https://aitools.flocci.in/tool/deepseek-models) — Free — DeepSeek V4-Pro and V4-Flash open-weight MoE models, MIT-licensed - [ERNIE 4.5 Open Models (Baidu)](https://aitools.flocci.in/tool/ernie-4-5-open-models-baidu) — Free — ERNIE-4.5 family including 21B-A3B Thinking and VL-424B-A47B multimodal MoE - [Falcon (TII)](https://aitools.flocci.in/tool/falcon-tii) — Free — Falcon-H1 hybrid Transformer-Mamba family for efficient long-context inference - [Hunyuan (Tencent)](https://aitools.flocci.in/tool/hunyuan-tencent) — Free — Open-weight Hunyuan Hy3 and Hy4-preview flagship text models - [MiniMax M2 (open weights)](https://aitools.flocci.in/tool/minimax-m2-open-weights) — Free — MiniMax's open-weight M-series: M3 multimodal MoE with very long context, plus M2.x agentic-coding checkpoints - [Mistral Open Models (Magistral, Devstral, Voxtral)](https://aitools.flocci.in/tool/mistral-open-models-magistral-devstral-voxtral) — Free — Open-weight Mistral Medium 3.5, Small 4 and Large 3 general models - [OLMo (Ai2)](https://aitools.flocci.in/tool/olmo-ai2) — Free — Only major model family shipping full pretraining data (Dolma), code and intermediate checkpoints, not just weights - [Qwen (Alibaba)](https://aitools.flocci.in/tool/qwen-alibaba) — Free — Open Qwen3.x family from phone-sized to multi-trillion-parameter MoE, Apache 2.0 - [Seed-OSS (ByteDance)](https://aitools.flocci.in/tool/seed-oss-bytedance) — Free — Seed-OSS-36B open reasoning LLM with native 512K-token context, Apache 2.0 - [Step (StepFun)](https://aitools.flocci.in/tool/step-stepfun) — Free — Step-3.7-Flash open multimodal model with FP8, NVFP4 and GGUF builds - [Xiaomi MiMo](https://aitools.flocci.in/tool/xiaomi-mimo) — Free — MiMo-V2.6-Pro (1T parameters) and V2.6-Flash (311B) open-weight reasoning and agent models - [EXAONE (LG AI Research)](https://aitools.flocci.in/tool/exaone-lg-ai-research) — Free — EXAONE 4.5: LG's first open-weight vision-language model for industrial use cases - [Gemma (Google)](https://aitools.flocci.in/tool/gemma-google) — Free — Google's lightweight open models (Gemma 4) - [GPT-OSS (OpenAI)](https://aitools.flocci.in/tool/gpt-oss-openai) — Free — OpenAI's Apache-2.0 open-weight reasoning models (120B and 20B) - [IBM Granite](https://aitools.flocci.in/tool/ibm-granite) — Free — Apache 2.0 licensed enterprise-grade model family - [Kimi K3 (Moonshot AI)](https://aitools.flocci.in/tool/kimi-k3-moonshot-ai) — Free — Open-weight multimodal MoE with about 2.8T total and 104B active parameters, 1M-token context - [Llama (Meta)](https://aitools.flocci.in/tool/llama-meta) — Free — Meta's flagship open-weight family (Llama 4) - [LongCat (Meituan)](https://aitools.flocci.in/tool/longcat-meituan) — Free — LongCat-2.0 open-weight 1.8T-parameter model with FP8 and INT8 builds - [Nemotron (NVIDIA)](https://aitools.flocci.in/tool/nemotron-nvidia) — Free — NVIDIA's open Nemotron 3 family of efficient agentic models - [Phi (Microsoft)](https://aitools.flocci.in/tool/phi-microsoft) — Free — MIT-licensed small language models for on-device and edge use - [Pleias](https://aitools.flocci.in/tool/pleias) — Free — Fully open-weight, open-dataset small language models (350M-3B) explicitly built for EU AI Act compliance - [Sarvam Open Models (30B / 105B)](https://aitools.flocci.in/tool/sarvam-open-models-30b-105b) — Free — Sarvam-105B and Sarvam-30B open models built for Indian languages, with FP8 and GGUF builds - [TimesFM (Google Research)](https://aitools.flocci.in/tool/timesfm-google-research) — Free — Pretrained decoder-only foundation model for time-series forecasting - [GLM (Z.ai)](https://aitools.flocci.in/tool/glm-z-ai) — Freemium — GLM-5.x flagship and Flash models released as open weights on Hugging Face --- Source: Flocci AI Tools — https://aitools.flocci.in/best/best-notable-open-models Facts last verified: October 2026