Baseten Integrations

DeepSeek-V3 is a state-of-the-art AI model designed to deliver unparalleled performance in natural language understanding, advanced reasoning, and decision-making tasks. Leveraging next-generation neural architectures, it integrates extensive datasets and fine-tuned algorithms to tackle complex challenges across diverse domains such as research, development, business intelligence, and automation. With a focus on scalability and efficiency, DeepSeek-V3 provides developers and enterprises with cutting-edge tools to accelerate innovation and achieve transformative outcomes.

DeepSeek-R1 is an advanced open-source reasoning model developed by DeepSeek, designed to rival OpenAI's Model o1. Accessible via web, app, and API, it excels in complex tasks such as mathematics and coding, demonstrating superior performance on benchmarks like the American Invitational Mathematics Examination (AIME) and MATH. DeepSeek-R1 employs a mixture of experts (MoE) architecture with 671 billion total parameters, activating 37 billion parameters per token, enabling efficient and accurate reasoning capabilities. This model is part of DeepSeek's commitment to advancing artificial general intelligence (AGI) through open-source innovation.

Tülu 3 is an advanced instruction-following language model developed by the Allen Institute for AI (Ai2), designed to enhance capabilities in areas such as knowledge, reasoning, mathematics, coding, and safety. Built upon the Llama 3 Base, Tülu 3 employs a comprehensive four-stage post-training process: meticulous prompt curation and synthesis, supervised fine-tuning on a diverse set of prompts and completions, preference tuning using both off- and on-policy data, and a novel reinforcement learning approach to bolster specific skills with verifiable rewards. This open-source model distinguishes itself by providing full transparency, including access to training data, code, and evaluation tools, thereby closing the performance gap between open and proprietary fine-tuning methods. Evaluations indicate that Tülu 3 outperforms other open-weight models of similar size, such as Llama 3.1-Instruct and Qwen2.5-Instruct, across various benchmarks.

The open source AI model you can fine-tune, distill and deploy anywhere. Our latest instruction-tuned model is available in 8B, 70B and 405B versions. Using our open ecosystem, build faster with a selection of differentiated product offerings to support your use cases. Choose from real-time inference or batch inference services. Download model weights to further optimize cost per token. Adapt for your application, improve with synthetic data and deploy on-prem or in the cloud. Use Llama system components and extend the model using zero shot tool use and RAG to build agentic behaviors. Leverage 405B high quality data to improve specialized models for specific use cases.

The open-source AI model you can fine-tune, distill and deploy anywhere is now available in more versions. Choose from 1B, 3B, 11B or 90B, or continue building with Llama 3.1. Llama 3.2 is a collection of large language models (LLMs) pretrained and fine-tuned in 1B and 3B sizes that are multilingual text only, and 11B and 90B sizes that take both text and image inputs and output text. Develop highly performative and efficient applications from our latest release. Use our 1B or 3B models for on device applications such as summarizing a discussion from your phone or calling on-device tools like calendar. Use our 11B or 90B models for image use cases such as transforming an existing image into something new or getting more information from an image of your surroundings.

Llama 3.3 is the latest iteration in the Llama series of language models, developed to push the boundaries of AI-powered understanding and communication. With enhanced contextual reasoning, improved language generation, and advanced fine-tuning capabilities, Llama 3.3 is designed to deliver highly accurate, human-like responses across diverse applications. This version features a larger training dataset, refined algorithms for nuanced comprehension, and reduced biases compared to its predecessors. Llama 3.3 excels in tasks such as natural language understanding, creative writing, technical explanation, and multilingual communication, making it an indispensable tool for businesses, developers, and researchers. Its modular architecture allows for customizable deployment in specialized domains, ensuring versatility and performance at scale.

​LiteLLM is a versatile platform designed to streamline interactions with over 100 Large Language Models (LLMs) through a unified interface. It offers both a Proxy Server (LLM Gateway) and a Python SDK, enabling developers to integrate various LLMs seamlessly into their applications. The Proxy Server facilitates centralized management, allowing for load balancing, cost tracking across projects, and consistent input/output formatting compatible with OpenAI standards. This setup supports multiple providers. It ensures robust observability by generating unique call IDs for each request, aiding in precise tracking and logging across systems. Developers can leverage pre-defined callbacks to log data using various tools. For enterprise users, LiteLLM offers advanced features like Single Sign-On (SSO), user management, and professional support through dedicated channels like Discord and Slack.

Llama 4 Maverick is one of the most advanced multimodal AI models from Meta, featuring 17 billion active parameters and 128 experts. It surpasses its competitors like GPT-4o and Gemini 2.0 Flash in a broad range of benchmarks, especially in tasks related to coding, reasoning, and multilingual capabilities. Llama 4 Maverick combines image and text understanding, enabling it to deliver industry-leading results in image-grounding tasks and precise, high-quality output. With its efficient performance at a reduced parameter size, Maverick offers exceptional value, especially in general assistant and chat applications.

Llama 4 Scout is a powerful 17 billion active parameter multimodal AI model that excels in both text and image processing. With an industry-leading context length of 10 million tokens, it outperforms its predecessors, including Llama 3, in tasks such as multi-document summarization and parsing large codebases. Llama 4 Scout is designed to handle complex reasoning tasks while maintaining high efficiency, making it perfect for use cases requiring long-context comprehension and image grounding. It offers cutting-edge performance in image-related tasks and is particularly well-suited for applications requiring both text and visual understanding.

Qwen3, the latest iteration of the Qwen family of large language models, introduces groundbreaking features that enhance performance across coding, math, and general capabilities. With models like the Qwen3-235B-A22B and Qwen3-30B-A3B, Qwen3 achieves impressive results compared to top-tier models, thanks to its hybrid thinking modes that allow users to control the balance between deep reasoning and quick responses. The platform supports 119 languages and dialects, making it an ideal choice for global applications. Its pre-training process, which uses 36 trillion tokens, enables robust performance, and advanced reinforcement learning (RL) techniques continue to refine its capabilities. Available on platforms like Hugging Face and ModelScope, Qwen3 offers a powerful tool for developers and researchers working in diverse fields.

Nomic Embed is a suite of open source, high-performance embedding models designed for various applications, including multilingual text, multimodal content, and code. The ecosystem includes models like Nomic Embed Text v2, which utilizes a Mixture-of-Experts (MoE) architecture to support over 100 languages with efficient inference using 305M active parameters. Nomic Embed Text v1.5 offers variable embedding dimensions (64 to 768) through Matryoshka Representation Learning, enabling developers to balance performance and storage needs. For multimodal applications, Nomic Embed Vision v1.5 aligns with the text models to provide a unified latent space for text and image data, facilitating seamless multimodal search. Additionally, Nomic Embed Code delivers state-of-the-art performance on code embedding tasks across multiple programming languages.

BGE (BAAI General Embedding) is a comprehensive retrieval toolkit designed for search and Retrieval-Augmented Generation (RAG) applications. It offers inference, evaluation, and fine-tuning capabilities for embedding models and rerankers, facilitating the development of advanced information retrieval systems. The toolkit includes components such as embedders and rerankers, which can be integrated into RAG pipelines to enhance search relevance and accuracy. BGE supports various retrieval methods, including dense retrieval, multi-vector retrieval, and sparse retrieval, providing flexibility to handle different data types and retrieval scenarios. The models are available through platforms like Hugging Face, and the toolkit provides tutorials and APIs to assist users in implementing and customizing their retrieval systems. By leveraging BGE, developers can build robust and efficient search solutions tailored to their specific needs.

ZenCtrl is an open source AI image generation toolkit developed by Fotographer AI, designed to produce high-quality, multi-view, and diverse-scene outputs from a single image without any training. It enables precise regeneration of objects and subjects from any angle and background, offering real-time element regeneration that provides both stability and flexibility in creative workflows. ZenCtrl allows users to regenerate subjects from any angle, swap backgrounds or clothing with just a click, and start generating results immediately without the need for additional training. By leveraging advanced image processing techniques, it ensures high accuracy without the need for extensive training data. The model's architecture is composed of lightweight sub-models, each fine-tuned on task-specific data to excel at a single job, resulting in a lean system that delivers sharper, more controllable results.

Over the last few weeks we all have been overwhelmed by the response and have been working hard to ensure a safe and ethical release, incorporating data from our beta model tests and community for the developers to act on. In cooperation with the tireless legal, ethics and technology teams at HuggingFace and amazing engineers at CoreWeave. We have developed an AI-based Safety Classifier included by default in the overall software package. This understands concepts and other factors in generations to remove outputs that may not be desired by the model user. The parameters of this can be readily adjusted and we welcome input from the community how to improve this. Image generation models are powerful, but still need to improve to understand how to represent what we want better.

Canopy Labs has introduced Orpheus, a family of state-of-the-art speech large language models (LLMs) designed for human-level speech generation. These models are built on the Llama-3 architecture and are trained on over 100,000 hours of English speech data, enabling them to produce natural intonation, emotion, and rhythm that surpasses current state-of-the-art closed source models. Orpheus supports zero-shot voice cloning, allowing users to replicate voices without prior fine-tuning, and offers guided emotion and intonation control through simple tags. The models achieve low latency, with approximately 200ms streaming latency for real-time applications, reducible to around 100ms with input streaming. Canopy Labs has released both pre-trained and fine-tuned 3B-parameter models under the permissive Apache 2.0 license, with plans to release smaller models of 1B, 400M, and 150M parameters for use on resource-constrained devices.

CAMB.AI's MARS6 is a groundbreaking text-to-speech (TTS) model that has become the first speech model accessible on Amazon Web Services (AWS) Bedrock platform. This integration allows developers to incorporate advanced TTS capabilities into generative AI applications, facilitating the creation of enhanced voice assistants, engaging audiobooks, interactive media, and various audio-centric experiences. MARS6's advanced algorithms enable natural and expressive speech synthesis, setting a new standard for TTS conversion. Developers can access MARS6 directly through the Amazon Bedrock platform, ensuring seamless integration into applications and enhancing user engagement and accessibility. The inclusion of MARS6 in AWS Bedrock's diverse selection of foundation models underscores CAMB.AI's commitment to advancing machine learning and artificial intelligence, providing developers with vital tools to create rich audio experiences supported by AWS's reliable and scalable infrastructure.

We’ve trained and are open-sourcing a neural net called Whisper that approaches human-level robustness and accuracy in English speech recognition. Whisper is an automatic speech recognition (ASR) system trained on 680,000 hours of multilingual and multitask supervised data collected from the web. We show that the use of such a large and diverse dataset leads to improved robustness to accents, background noise, and technical language. Moreover, it enables transcription in multiple languages, as well as translation from those languages into English. We are open-sourcing models and inference code to serve as a foundation for building useful applications and for further research on robust speech processing. The Whisper architecture is a simple end-to-end approach, implemented as an encoder-decoder Transformer. Input audio is split into 30-second chunks, converted into a log-Mel spectrogram, and then passed into an encoder.

Stable Diffusion XL or SDXL is the latest image generation model that is tailored towards more photorealistic outputs with more detailed imagery and composition compared to previous SD models, including SD 2.1. With Stable Diffusion XL you can now make more realistic images with improved face generation, produce legible text within images, and create more aesthetically pleasing art using shorter prompts.

Mixedbread is a fully-managed AI search engine that allows users to build production-ready AI search and Retrieval-Augmented Generation (RAG) applications. It offers a complete AI search stack, including vector stores, embedding and reranking models, and document parsing. Users can transform raw data into intelligent search experiences that power AI agents, chatbots, and knowledge systems without the complexity. It integrates with tools like Google Drive, SharePoint, Notion, and Slack. Its vector stores enable users to build production search engines in minutes, supporting over 100 languages. Mixedbread's embedding and reranking models have achieved over 50 million downloads and outperform OpenAI in semantic search and RAG tasks while remaining open-source and cost-effective. The document parser extracts text, tables, and layouts from PDFs, images, and complex documents, providing clean, AI-ready content without manual preprocessing.