Model Catalog

New MiMo-V2.6-Flash-RL is a 309B-parameter (15B active) omnimodal Mixture-of-Experts reasoning model with a 1M-token context window, built for coding, agentic, visual, audio, and cybersecurity workloads.

New MiMo-V2.6-Pro-RL is a 1.02T-parameter (42B active) flagship omnimodal Mixture-of-Experts reasoning model with a 1M-token context window, built for coding, agentic, visual, audio, and cybersecurity workloads.

New Laya is a multilingual, non-autoregressive System One decision model for calibrated routing, scoring, guardrails, moderation, and other low-latency decisions.

New DeepSeek V4.1 Flash is a Mixture-of-Experts model built for efficient reasoning, tool use, and agentic workloads.

New DeepSeek V4 Flash Vision Exp is an experimental multimodal MoE model built for efficient visual reasoning, tool use, and agentic workloads. It matches the performance of DeepSeek V4 Flash while heavily increasing multimodal agent capabilities.

New Qwen3.8-Flash-Next-FP8 is a native multimodal MoE model with 125B language-model parameters and 6B activated parameters, plus efficient n-gram embeddings and MTP for long-context agentic workloads.

New GLM 5.3 Flash is a native multimodal 320B-parameter MoE model with 18B activated parameters, combining sparse and linear attention for efficient long-context coding and agentic workloads.

New Qwen3.8-27B is a compact 27B dense vision-language model with native image and video understanding, flexible thinking control, and stronger coding, research, and long-horizon agentic performance. Context length is 262K natively and extensible to 1M tokens.

New DeepSeek V4 Pro 0813 is an updated release of the 1.6T-parameter MoE model with 49B activated parameters and a one-million-token context window for advanced long-context reasoning and agentic tasks.

New NVIDIA Nemotron 3.5 Lightning 30B A3B is a 30B-parameter sparse language model, provided in an NVIDIA-optimized NVFP4 format for efficient enterprise inference.

New NVIDIA Nemotron 3.5 Lightning 30B A3B is a 30B-parameter sparse language model, provided in BF16 format for enterprise inference.

New Muse Glimmer is a 30B-parameter causal language model, distilled from Muse Spark and built for autonomous local agents. It combines multimodal understanding, multi-step reasoning, reliable tool use, and failure recovery on consumer hardware.

New DeepSeek V4 Flash 0731 is an updated release of the 284B-parameter MoE model with 13B activated parameters, built for efficient long-context intelligence.

Inkling is a versatile multimodal Mixture-of-Experts transformer (975B total / 41B active) that accepts text, image, and audio inputs and generates text, targeting conversational, agentic, and coding workloads. This is the NVFP4-quantized release.

RedHatAI's gemma-4-31B-it-FP8-dynamic was obtained by quantizing the weights and activations of google/gemma-4-31B-it to FP8 data type using dynamic per-token quantization.

GLM 5.2 FP8 is the FP8 checkpoint of Z.ai's 753B-parameter flagship MoE model for long-horizon tasks, with a solid 1M-token context.

MiniMax-M3 is a native multimodal model with 1M context. It has ~428B parameters and ~23B activated parameters.

Nemotron 3 Ultra 550B A55B BF16 is an NVIDIA large language model for advanced reasoning, agentic tool use, and long-context conversational workloads.

Nemotron 3 Ultra 550B A55B NVFP4 is an NVIDIA-optimized release for advanced reasoning, agentic tool use, and long-context conversational workloads.

GLM 5.1 FP8 is the FP8 checkpoint of Z.ai's 754B-parameter flagship MoE model for agentic engineering and long-horizon coding.

Kimi K2.6 is a native multimodal 1T-parameter MoE model with 32B activated parameters, aimed at long-horizon coding, agentic workflows, and autonomous task orchestration.

MiniMax M2.7 is a 229B-parameter MoE model for software engineering, agentic tool use, productivity workflows, and long-context reasoning.

NVIDIA-optimized NVFP4 release of MiniMax M2.7, a 230B-parameter sparse MoE model for complex software engineering, agentic tool use, and productivity workflows.

NVIDIA-optimized NVFP4 release of Moonshot AI Kimi K2.6, a 1T-parameter native multimodal MoE model for agentic coding and autonomous workflows.

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows.

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8 is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows.

NVIDIA-optimized NVFP4 release of Moonshot AI Kimi K2.5: a large multimodal model (text, image, video) with up to 256k context, quantized with NVIDIA Model Optimizer for efficient inference.

Mistral Small 4 is a frontier-class multimodal model in the Mistral Small line, with strong multilingual, vision, and long-context performance.

Cohere speech transcription model with great performance for automatic speech recognition and long-form audio.

Reasoning-optimized sparse MoE model (~398B total, ~13B active per token) with native extended chain-of-thought and strong agentic / tool-calling performance.

Gemma 4 is Google's open multimodal family built on the same research stack as Gemini. This instruction-tuned 31B variant supports image-text input and conversational use.

Gemma 4 26B A4B is a mixture-of-experts instruction-tuned model in the Gemma 4 line, with multimodal (image-text) conversational capabilities.

Compact Gemma 4 E4B instruction-tuned variant with multimodal (image-text and broader modality) support for efficient deployment.

Smallest Gemma 4 E2B instruction-tuned variant with multimodal (image-text and broader modality) support for edge-friendly deployment.

Qwen3.5 397B A17B is the largest of the next-generation multimodal models that combines unified vision-language learning, efficient hybrid architecture, large-scale reinforcement learning, and broad multilingual support to deliver highly capable, scalable, and globally accessible AI.

Qwen3-Coder-Next is an open-weight coding-focused model that delivers high performance with minimal active parameters, excels in agentic reasoning and tool use, and integrates seamlessly into real-world IDEs with long context support.

Nemotron-3-Super-120B-A12B-BF16 is a large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities.

Nemotron-3-Nano-30B-A3B-BF16 is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks.

The largest model in the Ministral 3 family, Ministral 3 14B offers frontier capabilities and performance comparable to its larger Mistral Small 3.2 24B counterpart. A powerful and efficient language model with vision capabilities.

Mistral Large 3 is a state-of-the-art general-purpose Multimodal granular Mixture-of-Experts model with 41B active parameters and 675B total parameters trained from the ground up with 3000 H200s.

GPT OSS is the latest OpenAI open-model designed for powerful reasoning, agentic tasks, and versatile developer use cases. GPT OSS 120B is intended for production, general purpose, and high reasoning use cases.

GPT OSS is the latest OpenAI open-model designed for powerful reasoning, agentic tasks, and versatile developer use cases. GPT OSS 20B is intended for lower latency, and local or specialized use cases.