Models

Reasoning-focused model with chain-of-thought capabilities rivaling o1.

Updated R1 with improved reasoning accuracy and reduced hallucination.

Open-weight 671B MoE model with strong coding and reasoning at low cost.

Updated DeepSeek V3 with improved coding and reasoning performance.

Latest DeepSeek V3 with improved reasoning and coding. 671B MoE (37B active), MIT licensed, 164K context.

Strong reasoning and coding with 106B total, 12B active MoE architecture.

Open-source frontier model with 355B parameters. MIT licensed.

Frontier 744B model trained on Huawei Ascend chips. Open source with strong agentic capabilities.

Open-weight 117B MoE model (5.1B active) achieving near o4-mini reasoning. Apache 2.0 licensed, runs on a single 80GB GPU.

Mid-size open-weight Gemma model with vision support.

Largest Gemma 3 model with strong reasoning and instruction following.

Compact open-weight model for edge and mobile deployment.

Latest Gemma generation optimized for reasoning and agentic workflows.

Most capable open Gemma model with best intelligence-per-parameter.

State-of-the-art 1T MoE model with 32B active parameters. Strong coding and agentic capabilities.

Open-weight multimodal model with agent swarm mode supporting up to 100 parallel sub-agents.

Widely deployed open-weight model with strong general capabilities.

Largest open Llama 4 with 128 experts. 400B total, 17B active. Beats GPT-4o on benchmarks.

Natively multimodal MoE model with 10M context. 109B total, 17B active. Fits single H100.

Edge-optimized model with vision support. Apache 2.0 licensed.

Mistral's most capable model. 675B MoE with 41B active parameters.

Unified model combining fast instruct, deep reasoning, and multimodal chat. 119B params.

Largest Qwen 3 model with hybrid thinking modes for flexible reasoning control.

Mid-size Qwen 3 with strong coding and math capabilities. Open weight.

Compact Qwen 3 for edge and single-GPU deployment. Open weight.

Large Qwen 3.5 MoE model with 122B total, 10B active parameters.

Mid-size Qwen 3.5 MoE model with 35B total, 3B active parameters.

Largest Qwen 3.5 MoE model with 397B total, 17B active parameters.

Native multimodal Qwen with text, image, and video processing.

Compact Qwen 3.5 for single-GPU deployment.