Model Cards

First open-source transformer-based multilingual NMT model supporting high-quality translation across all 22 scheduled Indic languages.

Southeast Asian Languages In One Network — LLMs pretrained and instruct-tuned for the Southeast Asia region.

General embedding base model transforming text into 768-dimensional vectors.

General embedding large model transforming text into 1024-dimensional vectors.

Multi-functionality, multi-linguality, and multi-granularity embedding model.

Reranker model that takes question and document as input and outputs a relevance similarity score.

General embedding small model transforming text into 384-dimensional vectors.

12 billion parameter rectified flow transformer capable of generating images from text descriptions.

Image model capable of generating highly realistic and detailed images with multi-reference support.

Ultra-fast distilled image model delivering state-of-the-art quality for interactive workflows and real-time previews.

Ultra-fast distilled image model with enhanced quality; unifies image generation and editing in a single model.

Lightning-fast text-to-image generation model capable of producing high-quality 1024px images in a few steps.

Context-aware text-to-speech model applying natural pacing and expressiveness.

Context-aware English text-to-speech model that applies natural pacing, expressiveness, and fillers based on context.

Context-aware Spanish text-to-speech model that applies natural pacing, expressiveness, and fillers based on context.

First conversational speech recognition model built specifically for voice agents.

Deepgram speech-to-text model for transcribing audio.

SQL-specialized model designed to help non-technical users understand and query data in SQL databases.

300M parameter state-of-the-art open embedding model producing vector representations for search and retrieval; trained with data in 100+ languages.

Gemma 2B base model dedicated for inference with LoRA adapters.

128K context window model with multilingual support in over 140 languages; handles text and image input.

Google's most intelligent family of open models, derived from Gemini 3 research.

Lightweight, state-of-the-art open model built from the same research and technology as Gemini models.

Gemma 7B base model dedicated for inference with LoRA adapters.

Distilled BERT model fine-tuned on SST-2 for sentiment classification.

Industry-leading results in agentic tasks including instruction following and function calling; suited for RAG, multi-agent workflows, and edge deployments.

Most adaptable and prompt-responsive model with strengths in sharp graphic design, full-HD renders, and accurate text rendering.

Generates images with exceptional prompt adherence and coherent text rendering.

Open-source multimodal chatbot trained by fine-tuning LLaMA/Vicuna on visual instruction data; supports image captioning and visual question answering.

Stable Diffusion model fine-tuned to be better at photorealism.

Distilled from DeepSeek-R1 based on Qwen2.5; outperforms OpenAI o1-mini on several benchmarks.

DETection TRansformer model trained end-to-end on COCO 2017 object detection dataset (118k annotated images).

Full precision (fp16) generative text model with 7 billion parameters.

Llama 2 base model dedicated for inference with LoRA adapters.

Quantized (int8) generative text model with 7 billion parameters.

State-of-the-art performance on industry benchmarks with improved reasoning capabilities.

Quantized (int4) generative text model with 8 billion parameters.

Large multilingual model optimized for multilingual dialogue use cases.

Multilingual large language model optimized for multilingual dialogue use cases.

Quantized (int4) generative text model with 8 billion parameters.

Fast variant of Llama 3.1 8B optimized for multilingual dialogue use cases.

Llama 3.1 8B quantized to FP8 precision.

Multimodal model optimized for visual recognition, image reasoning, and captioning.

Lightweight model optimized for multilingual dialogue use cases.

Lightweight model optimized for multilingual dialogue including agentic retrieval and summarization.

Llama 3.3 70B quantized to fp8 precision and optimized for faster inference.

17 billion parameter natively multimodal model with 16 experts (mixture-of-experts architecture).

Llama 3.1 8B fine-tuned for content safety classification in both LLM inputs and responses.

Multilingual encoder-decoder model trained for many-to-many multilingual translation.

State-of-the-art performance on a wide range of industry benchmarks.

Transformer-based model with next-word prediction objective, trained on 1.4T tokens from web and synthetic sources.

50 layers deep image classification CNN trained on more than 1 million images from the ImageNet dataset.

Instruct fine-tuned version of the Mistral-7B model.

32k context window instruct model with updated rope-theta and no sliding-window attention.

Mistral 7B instruct v0.2 dedicated for inference with LoRA adapters.

State-of-the-art vision understanding with long context capabilities up to 128k tokens.

Frontier-scale open-source model with a 256k context window.

Frontier-scale open-source 1T parameter model with a 262.1k context window, designed for agentic workloads.

High-quality multi-lingual text-to-speech library.

Upgraded, retrained version of Nous Hermes 2 with function calling and JSON mode capabilities.

Hybrid MoE model with leading accuracy for multi-agent applications.

Open-weight model designed for production, general-purpose, high-reasoning use cases.

Open-weight model optimized for lower latency and local or specialized use cases.

General-purpose speech recognition model supporting multilingual recognition, speech translation, and language identification.

Pre-trained model for automatic speech recognition (ASR) and speech translation.

English-only version of the Whisper Tiny model trained on speech recognition.

Japanese text embedding model that converts Japanese text input into numerical vectors for information retrieval, text classification, and clustering.

Open source community-driven native audio turn detection model in its second version.

Code-specific large language model supporting six mainstream model sizes from 0.5B to 32B parameters.

Latest generation large language model with groundbreaking advancements in reasoning, instruction-following, and agent capabilities.

Latest Qwen family model specifically designed for text embedding and ranking.

Medium-sized reasoning model capable of thinking deeply for hard problems; competitive with DeepSeek-R1 and o1-mini.

Generate a new image from an input image using Stable Diffusion.

Stable Diffusion model with inpainting capability using a mask to selectively edit image regions.

Diffusion-based text-to-image model that generates and modifies images based on text prompts.

Small generative vision-language model primarily designed for image captioning and visual question answering.

Fast and efficient multilingual text generation model with a 131,072 token context window supporting 100+ languages.