Llama (Meta)
Notable Open Models
- Meta's flagship open-weight family (Llama 4)
- Multimodal and long-context variants
- Run locally or via any host
- Free, permissive license
AI Models & Local Execution (24)
24 free open-weight AI models and 0 more — pricing tier and real capabilities on every entry. No ads, no sponsored placement.
AllLocal LLM RunnersModel Hosting & Inference APIsNotable Open ModelsSpeech & Transcription EnginesFine-Tuning & Training FrameworksLLM Eval & ObservabilityLeaderboards & BenchmarksRobotics & Embodied AI
free
Qwen (Alibaba)
Notable Open Models
- Open Qwen3.x family from phone-sized to multi-trillion-parameter MoE, Apache 2.0
- Qwen3.6 mid-size models suit agentic coding on one high-memory machine
- Small Qwen3.5 sizes run on phones and laptops
- Free weights on Hugging Face and ModelScope
DeepSeek Models
Notable Open Models
- DeepSeek V4-Pro and V4-Flash open-weight MoE models, MIT-licensed
- Very long context with reasoning and non-reasoning modes
- Successors to the V3 and R1 lines
- Free weights plus free web chat
Gemma (Google)
Notable Open Models
- Google's lightweight open models (Gemma 4)
- Runs in as little as 16GB RAM
- On-device and multimodal variants
- Free and open
GPT-OSS (OpenAI)
Notable Open Models
- OpenAI's Apache-2.0 open-weight reasoning models (120B and 20B)
- Run locally or self-host, with configurable reasoning effort
- Safeguard variants for policy-based content classification
Pleias
Notable Open Models
- Fully open-weight, open-dataset small language models (350M-3B) explicitly built for EU AI Act compliance
- Trained exclusively on the 2-trillion-token open Common Corpus, not scraped/licensed text
- Purpose-built RAG models (Pleias-RAG-350M/1B) with built-in citation tracking, Apache 2.0 licensed
Mistral Open Models (Magistral, Devstral, Voxtral)
Notable Open Models
- Open-weight Mistral Medium 3.5, Small 4 and Large 3 general models
- Devstral for coding agents and Magistral for visible reasoning
- Voxtral open-weights speech models for transcription and voice agents
GLM (Z.ai)
Notable Open Models
- GLM-5.x flagship and Flash models released as open weights on Hugging Face
- GLM-4.x Flash text and vision models are free via the Z.ai API
- Strong agentic and coding results among open Chinese models
IBM Granite
Notable Open Models
- Apache 2.0 licensed enterprise-grade model family
- Hybrid Mamba/transformer architecture (Granite 4) for lower memory footprint
- Built-in focus on governance, provenance and enterprise compliance
OLMo (Ai2)
Notable Open Models
- Only major model family shipping full pretraining data (Dolma), code and intermediate checkpoints, not just weights
- Olmo 3 ships Base, Think, and Instruct variants at 7B/32B
- OlmoCore training framework and Open Instruct pipeline are also open source
Hunyuan (Tencent)
Notable Open Models
- Open-weight Hunyuan Hy3 and Hy4-preview flagship text models
- Hunyuan3D line for fast image-to-3D generation
- Hy-MT translation models covering dozens of language pairs
Falcon (TII)
Notable Open Models
- Falcon-H1 hybrid Transformer-Mamba family for efficient long-context inference
- Falcon-H1R small reasoning model with strong math results for its size
- Free commercial use under TII's Falcon license
EXAONE (LG AI Research)
Notable Open Models
- EXAONE 4.5: LG's first open-weight vision-language model for industrial use cases
- 33B flagship plus FP8/AWQ quantized variants for efficient local deployment
- Backed by LG's enterprise/manufacturing domain tuning
MiniMax M2 (open weights)
Notable Open Models
- MiniMax's open-weight M-series: M3 multimodal MoE with very long context, plus M2.x agentic-coding checkpoints
- Distinct from MiniMax's Hailuo video product
- Free weights on Hugging Face
Phi (Microsoft)
Notable Open Models
- MIT-licensed small language models for on-device and edge use
- Phi-4 reasoning and reasoning-vision variants decide when to think step by step
- Phi-4-mini and multimodal variants handle text, vision and audio
- Available on Hugging Face, Azure AI Foundry and Ollama
Nemotron (NVIDIA)
Notable Open Models
- NVIDIA's open Nemotron 3 family of efficient agentic models
- Top-ranked open embedding models (Llama-Embed-Nemotron)
- Open training datasets released alongside the models
- Pretrained decoder-only foundation model for time-series forecasting
- Zero-shot forecasts with quantile outputs
- Weights on Hugging Face, also in BigQuery
- Apache-2.0 licensed
- Open-weight multimodal MoE with about 2.8T total and 104B active parameters, 1M-token context
- Native text, image and video input with configurable thinking effort
- Earlier open Kimi K2.x (K2.5, K2.6, K2.7-Code) remain downloadable from the same org
- Weights free to download; also usable free in kimi.com chat with limits
- MiMo-V2.6-Pro (1T parameters) and V2.6-Flash (311B) open-weight reasoning and agent models
- Distilled MiMo-V2.6-Distill-Qwen-9B for single-GPU use
- V2-Flash, V2.5 and V2.5-Pro checkpoints also published
- LongCat-2.0 open-weight 1.8T-parameter model with FP8 and INT8 builds
- LongCat-Flash-Lite-Sparse 69B, LongCat-Flash-Prover and LongCat-Next any-to-any models
- Also AudioDiT speech and Video-Avatar models, all downloadable
- Step-3.7-Flash open multimodal model with FP8, NVFP4 and GGUF builds
- Step-3.5-Flash text model widely run locally
- 51 open models incl. audio and image models
- Sarvam-105B and Sarvam-30B open models built for Indian languages, with FP8 and GGUF builds
- Sarvam-Translate 4B for multilingual translation
- Downloadable from Hugging Face at no cost
- ERNIE-4.5 family including 21B-A3B Thinking and VL-424B-A47B multimodal MoE
- ERNIE-Image and ERNIE-Image-Turbo 8B text-to-image models
- Unlimited-OCR 3B document model (1.5M+ downloads)
- Seed-OSS-36B open reasoning LLM with native 512K-token context, Apache 2.0
- Base, Base-without-synthetic-data and Instruct checkpoints
- Adjustable thinking budget for reasoning length
Notable Open Models — straight answers
What are the best free open-weight AI models?
Flocci AI Tools lists 24 open-weight AI models, ordered free tiers first: DeepSeek Models, ERNIE 4.5 Open Models (Baidu), Falcon (TII), Hunyuan (Tencent) and MiniMax M2 (open weights). All of them are free or have a free tier. Every entry states what it uniquely does, not a score.
Are there completely free open-weight AI models?
Yes — 23 of these are marked fully free with no paid plan attached: DeepSeek Models, ERNIE 4.5 Open Models (Baidu), Falcon (TII), Hunyuan (Tencent) and MiniMax M2 (open weights). The rest are freemium, trial or paid, and each row states which.
What should I look for in open-weight AI models?
Start with what you need to do: download weights you can inspect, fine-tune and self-host. Then check the pricing tier, because "free" here means no paid plan at all while "freemium" means a free tier under paid plans. Every tool on this page lists both.
See the full list →Head-to-head comparisons
DeepSeek Models vs ERNIE 4.5 Open Models (Baidu)DeepSeek Models vs Falcon (TII)DeepSeek Models vs Hunyuan (Tencent)DeepSeek Models vs MiniMax M2 (open weights)DeepSeek Models vs Mistral Open Models (Magistral, Devstral, Voxtral)ERNIE 4.5 Open Models (Baidu) vs Falcon (TII)ERNIE 4.5 Open Models (Baidu) vs Hunyuan (Tencent)ERNIE 4.5 Open Models (Baidu) vs MiniMax M2 (open weights)
Free alternatives
Llama (Meta) alternativesQwen (Alibaba) alternativesDeepSeek Models alternativesGemma (Google) alternativesGPT-OSS (OpenAI) alternativesPleias alternativesMistral Open Models (Magistral, Devstral, Voxtral) alternativesGLM (Z.ai) alternatives
Looking for a ranked shortlist instead? See the best free open-weight AI models.