Open Source AI Models — 527 Open Weight LLMs Compared

Contents

🏆 Flagship Open-Weight Models

The most capable open-weight models available today, from leading AI labs:

Model	Provider	Context	Tool Call	Reasoning	Price (in/out per 1M)
`llama-4-maverick`	Meta	1M	✓	✗	Varies
`llama-4-scout`	Meta	10M	✓	✗	Varies
`deepseek-r1`	DeepSeek	128K	✓	✓	Varies
`deepseek-v3`	DeepSeek	128K	✓	✗	Varies
`qwen3-235b-a22b`	Alibaba	128K	✓	✓	Varies
`qwen3-32b`	Alibaba	128K	✓	✓	Varies
`llama-3.3-70b-instruct`	Meta	128K	✓	✗	Varies
`gemma-3-27b-it`	Google	128K	✓	✗	Free
`phi-4`	Microsoft	16K	✓	✗	Varies
`command-a`	Cohere	256K	✓	✗	Varies
`mistral-large-2411`	Mistral	128K	✓	✗	Varies

81 open-weight models you can use for free through their provider APIs. These are ideal for prototyping, testing, and learning:

Model	Provider	Context	Tool Call	Reasoning
`gemma-3-27b-it`	Google	128K	✓	✗
`gemma-3-12b-it`	Google	128K	✓	✗
`gemma-3-4b-it`	Google	128K	✓	✗
`gemma-3-1b-it`	Google	128K	✗	✗
`qwen3-235b-a22b`	Alibaba	128K	✓	✓
`qwen3-30b-a3b`	Alibaba	128K	✓	✓
`qwen3-32b`	Alibaba	128K	✓	✓
`qwen3-14b`	Alibaba	128K	✓	✓
`qwen3-8b`	Alibaba	128K	✓	✓
`qwen3-4b`	Alibaba	128K	✓	✓
`qwen3-1.7b`	Alibaba	128K	✓	✓
`qwen3-0.6b`	Alibaba	128K	✓	✓
`llama-4-maverick`	Meta	1M	✓	✗
`llama-4-scout`	Meta	10M	✓	✗
`llama-3.3-70b-instruct`	Meta	128K	✓	✗

→ See all 81 free AI models (including non-open-weight)

375 open-weight models support tool/function calling — essential for AI agents and agentic workflows:

Llama 4 Maverick/Scout — Meta's latest with native tool calling, 1M–10M context
Qwen3 series — All sizes support tool calling + reasoning (0.6B to 235B)
DeepSeek R1/V3 — Strong tool calling with 128K context
Gemma 3 (1B–27B) — Google's lightweight models with tool calling
Command A — Cohere's 111B model optimized for enterprise tool use
Mistral Large — 123B parameter model with robust function calling

231 open-weight models with reasoning capabilities — these can "think step by step" for complex tasks:

Model	Provider	Context	Tool Call	Key Strength
`deepseek-r1`	DeepSeek	128K	✓	Best open-weight reasoning, rivals o1
`qwen3-235b-a22b`	Alibaba	128K	✓	MoE architecture, thinking mode
`qwen3-32b`	Alibaba	128K	✓	Dense reasoning, strong benchmarks
`qwen3-30b-a3b`	Alibaba	128K	✓	Lightweight MoE reasoning
`qwen3-14b`	Alibaba	128K	✓	Mid-size reasoning model
`qwen3-8b`	Alibaba	128K	✓	Small but capable reasoning

269 open-weight models can process images alongside text — useful for document analysis, visual Q&A, and multimodal applications:

Llama 4 Maverick/Scout — Native multimodal with 1M–10M context, process images + text
Qwen3 series — Vision-capable across all sizes
Gemma 3 (1B–27B) — Google's vision-language models, free to use
DeepSeek R1/V3 — Reasoning + vision capabilities
Command A — Enterprise-grade vision + tool calling

Open-weight models with the largest context windows — essential for processing long documents, codebases, and multi-turn conversations:

Model	Provider	Context Window	Tool Call	Reasoning
`llama-4-scout`	Meta	10M	✓	✗
`llama-4-maverick`	Meta	1M	✓	✗
`command-a`	Cohere	256K	✓	✗
`deepseek-r1`	DeepSeek	128K	✓	✓
`deepseek-v3`	DeepSeek	128K	✓	✗
`qwen3-235b-a22b`	Alibaba	128K	✓	✓
`llama-3.3-70b-instruct`	Meta	128K	✓	✗
`gemma-3-27b-it`	Google	128K	✓	✗
`mistral-large-2411`	Mistral	128K	✓	✗

Need free API access? → Start with gemma-3-27b-it (Google, free) or qwen3-32b (Alibaba, free)
Building AI agents? → llama-4-maverick (1M context + tool calling) or deepseek-r1 (reasoning + tools)
Processing long documents? → llama-4-scout (10M context) or llama-4-maverick (1M context)
Complex reasoning tasks? → deepseek-r1 (best open-weight reasoning) or qwen3-235b-a22b
Vision/image understanding? → llama-4-maverick or gemma-3-27b-it
Edge/mobile deployment? → qwen3-0.6b or gemma-3-1b-it (smallest open-weight)
Enterprise tool use? → command-a (256K context, optimized for RAG + tools)