DeepSeek
DeepSeek V4
DeepSeek's open-weight V4 family and hosted API releases, including the current V4 Pro production build.
OpenRouter-informed model coverage
Open-Source Models
Open-weight and open-source model families that developers actively compare for local and hosted use. Availability and licensing are checked against each model’s primary source.
Alibaba
Qwen3
A multilingual open-weight family spanning dense, MoE, reasoning and vision-language variants.
Meta
Llama 4
Meta’s multimodal open-weight Llama family, offered in Scout and Maverick variants.
Mistral AI
Mistral Small 3.1
A compact multimodal open model for practical local deployment and self-hosted applications.
Gemma 3
Google’s lightweight open-weight family for text and vision tasks across a range of sizes.
OpenAI
gpt-oss
Open-weight reasoning models designed for local or hosted deployment, including 20B and 120B variants.
Ai2
OLMo 2
Fully open language models accompanied by transparent training data, code and evaluation materials.
Microsoft
Phi-4
Microsoft’s small-model family, oriented toward efficient reasoning and on-device or constrained deployments.
NVIDIA
Nemotron
NVIDIA’s open-weight model line for reasoning, agentic workloads and efficient inference.
Moonshot AI
Kimi K2
An open-weight mixture-of-experts model family focused on coding, tool use and agentic work.