Model Catalog

DeepSeek V4 Flash 0731 is a 284B Mixture-of-Experts model built for coding, tool use, and agentic workflows, with 13B active parameters and a 1M-token context window.

Kimi K3 is Moonshot AI’s 2.8T-parameter flagship model for long-horizon agentic work, with native vision, always-on reasoning, and a 1M-token context window.

Laguna S 2.1 is Poolside’s 118B Mixture-of-Experts model for agentic coding and long-horizon tool use, with 8B active parameters and a context window of up to 1M tokens.

Bonsai 27B is PrismML’s family of 1-bit and ternary Qwen3.6 27B models, with vision, reasoning, tool use, and a 262K-token context window.

GLM-5.2 is Z.ai’s 753B Mixture-of-Experts flagship model for long-horizon coding and agentic work, with 40B active parameters and flexible reasoning effort.

Granite 4.1 models are new and improved granite models which have gone through an improved post-training pipeline, including supervised finetuning and reinforcement learning alignment, resulting in enhanced tool calling, instruction following, and chat capabilities.

NVIDIA Nemotron 3 Nano Omni is an open multimodal model with highest efficiency that powers sub-agents to complete tasks faster across vision, audio, and language

Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.

Gemma 4 is Google's most capable family of open models, built from Gemini 3 research. Supports vision input and available in multiple sizes for on-device deployment.

NVIDIA Nemotron 3 Super, a 120B open hybrid MoE model (12B active), supporting up to 1M tokens context window

Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency

LFM2 is a family of hybrid models designed for on-device deployment. LFM2-24B-A2B is the largest model in the family, scaling the architecture to 24 billion parameters while keeping inference efficient.

Qwen3 Coder Next is an 80B MoE with 3B active parameters designed for coding agents and local development. Excels at long-horizon reasoning, complex tool usage, and recovery from execution failures.

Open source coding models by Z.ai, based on a new base model and specializing in coding and tool calling.

FunctionGemma is a lightweight, open model from Google, built as a foundation for creating your own specialized function calling models.

General purpose reasoning and chat model trained from scratch by NVIDIA. Contains 30B total parameters with only 3.5B active at a time for low-latency MoE inference

GLM 4.6V Flash is a 9B vision-language model optimized for local deployment and low-latency applications.

Second-generation Devstral for agentic coding. Built for tool use to explore codebases, edit multiple files, and power software engineering agents with newly added vision support.

Rnj-1 is a family of 8B parameter open-weight, dense models trained from scratch by Essential AI.

Ministral 3 series, available in three model sizes: 3B, 8B, and 14B parameters. Provides best of class cost-to-performance ratio.

Hybrid attention architecture, high-sparsity Mixture-of-Experts 80B model (active 3B).

Olmo 3 is a family of Open language models designed to enable the science of language models.

The olmOCR 2 model is a Vision Language Model (VLM) from Allen AI.

MiniMax M2 is a 230B MoE (10B active) model built for coding and agentic workflows

gpt-oss-safeguard-20b and gpt-oss-safeguard-120b are open safety models from OpenAI, building on gpt-oss. Trained to help classify text content based on customizable policies.

Qwen's latest vision-language model. Includes comprehensive upgrades to visual perception, spatial reasoning, and image understanding.

Granite 4.0 language models are lightweight, state-of-the-art open models that natively support multilingual capabilities, coding tasks, RAG, tool use, and JSON output.

Advanced reasoning model from ByteDance with flexible "thinking budget" control and ability to reflect on the length of its own reasoning

The latest version of the Qwen3 model family, featuring 4B, 30B, and 235B dense and MoE models, both thinking and non-thinking variants.

OpenAI's first open source LLM. Comes in 2 sizes: 20B and 120B. Supports configurable reasoning effort (low, medium, high). Trained for tool use. Apache 2.0 licensed.

State-of-the-art, Mixture-of-Experts local coding model with native support for 256K context length. Available in 30B (3B active) and 480B (35B active) sizes.

Medium-size Mixture-of-Experts model from Baidu's new Ernie 4.5 line of foundation models.

LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency.

Devstral is a coding model from Mistral AI. It excels at using tools to explore codebases, editing multiple files and power software engineering agents.

Gemma 3n is a generative AI model optimized for use in everyday devices, such as phones, laptops, and tablets.

Mistrall Small is a 'knowledge-dense' 24B multi-modal (image input) local model that supports up to 128 token context length.

MistralAI's open-weight reasoning model. 24B dense transformer model supporting up to 128K token context window. The model is capable of long chains of reasoning traces before providing answers.

General purpose dense transformer designed for multilingual use cases. Built in collaboration between MistralAI and NVIDIA.

Qwen2.5-VL is a performant vision-language model, capable of recognizing common objects and text. Supports context length of 128k tokens in a variety of human languages.

State-of-the-art image + text input models from Google, built from the same research and tech used to create the Gemini models

Phi-4-mini-reasoning is a lightweight open model built upon synthetic data with a focus on high-quality, reasoning dense data.

phi-4 is a state-of-the-art open model built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets.

Mistral AI's latest coding model, Codestral can handle both instructions and code completions with ease in over 80 programming languages.

One of the most popular open-source LLMs, Mistral's 7B Instruct model's balance of speed, size, and performance makes it a great general-purpose daily driver.

The first batch of Qwen3 models (Qwen3-2504), a collection of dense and MoE models ranging from 4B to 235B. These are general purpose models that score highly on benchmarks.

Distilled version of the DeepSeek-R1-0528 model, created by continuing the post-training process on the Qwen3 8B Base model using Chain-of-Thought (CoT) from DeepSeek-R1-0528.