Baseten Integrations

DeepSeek-V3 is a remarkable leap forward in the realm of artificial intelligence, meticulously crafted to demonstrate exceptional prowess in understanding natural language, complex reasoning, and effective decision-making. By leveraging cutting-edge neural network architectures, this model assimilates extensive datasets along with sophisticated algorithms to tackle challenging issues in numerous domains such as research, development, business analytics, and automation. With a strong emphasis on scalability and operational efficiency, DeepSeek-V3 provides developers and organizations with groundbreaking tools that can greatly accelerate advancements and yield transformative outcomes. Additionally, its adaptability ensures that it can be applied in a multitude of contexts, thereby enhancing its significance across various sectors. This innovative approach not only streamlines processes but also opens new avenues for exploration and growth in artificial intelligence applications.

DeepSeek-R1 represents a state-of-the-art open-source reasoning model developed by DeepSeek, designed to rival OpenAI's Model o1. Accessible through web, app, and API platforms, it demonstrates exceptional skills in intricate tasks such as mathematics and programming, achieving notable success on exams like the American Invitational Mathematics Examination (AIME) and MATH. This model employs a mixture of experts (MoE) architecture, featuring an astonishing 671 billion parameters, of which 37 billion are activated for every token, enabling both efficient and accurate reasoning capabilities. As part of DeepSeek's commitment to advancing artificial general intelligence (AGI), this model highlights the significance of open-source innovation in the realm of AI. Additionally, its sophisticated features have the potential to transform our methodologies in tackling complex challenges across a variety of fields, paving the way for novel solutions and advancements. The influence of DeepSeek-R1 may lead to a new era in how we understand and utilize AI for problem-solving.

Tülu 3 represents a state-of-the-art language model designed by the Allen Institute for AI (Ai2) with the objective of enhancing expertise in various domains such as knowledge, reasoning, mathematics, coding, and safety. Built on the foundation of the Llama 3 Base, it undergoes an intricate four-phase post-training process: meticulous prompt curation and synthesis, supervised fine-tuning across a diverse range of prompts and outputs, preference tuning with both off-policy and on-policy data, and a distinctive reinforcement learning approach that bolsters specific skills through quantifiable rewards. This open-source model is distinguished by its commitment to transparency, providing comprehensive access to its training data, coding resources, and evaluation metrics, thus helping to reduce the performance gap typically seen between open-source and proprietary fine-tuning methodologies. Performance evaluations indicate that Tülu 3 excels beyond similarly sized models, such as Llama 3.1-Instruct and Qwen2.5-Instruct, across multiple benchmarks, emphasizing its superior effectiveness. The ongoing evolution of Tülu 3 not only underscores a dedication to enhancing AI capabilities but also fosters an inclusive and transparent technological landscape. As such, it paves the way for future advancements in artificial intelligence that prioritize collaboration and accessibility for all users.

We are excited to unveil an open-source AI model that offers the ability to be fine-tuned, distilled, and deployed across a wide range of platforms. Our latest instruction-tuned model is available in three different sizes: 8B, 70B, and 405B, allowing you to select an option that best fits your unique needs. The open ecosystem we provide accelerates your development journey with a variety of customized product offerings tailored to meet your specific project requirements. You can choose between real-time inference and batch inference services, depending on what your project requires, giving you added flexibility to optimize performance. Furthermore, downloading model weights can significantly enhance cost efficiency per token while you fine-tune the model for your application. To further improve performance, you can leverage synthetic data and seamlessly deploy your solutions either on-premises or in the cloud. By taking advantage of Llama system components, you can also expand the model's capabilities through the use of zero-shot tools and retrieval-augmented generation (RAG), promoting more agentic behaviors in your applications. Utilizing the extensive 405B high-quality data enables you to fine-tune specialized models that cater specifically to various use cases, ensuring that your applications function at their best. In conclusion, this empowers developers to craft innovative solutions that not only meet efficiency standards but also drive effectiveness in their respective domains, leading to a significant impact on the technology landscape.

The newest version of the open-source AI framework, which can be customized and utilized across different platforms, is available in several configurations: 1B, 3B, 11B, and 90B, while still offering the option to use Llama 3.1. Llama 3.2 includes a selection of large language models (LLMs) that are pretrained and fine-tuned specifically for multilingual text processing in 1B and 3B sizes, whereas the 11B and 90B models support both text and image inputs, generating text outputs. This latest release empowers users to build highly effective applications that cater to specific requirements. For applications running directly on devices, such as summarizing conversations or managing calendars, the 1B or 3B models are excellent selections. On the other hand, the 11B and 90B models are particularly suited for tasks involving images, allowing users to manipulate existing pictures or glean further insights from images in their surroundings. Ultimately, this broad spectrum of models opens the door for developers to experiment with creative applications across a wide array of fields, enhancing the potential for innovation and impact.

The latest iteration in the Llama series, Llama 3.3, marks a notable leap forward in the realm of language models, designed to improve AI's abilities in both understanding and communication. It features enhanced contextual reasoning, more refined language generation, and state-of-the-art fine-tuning capabilities that yield remarkably accurate, human-like responses for a wide array of applications. This version benefits from a broader training dataset, advanced algorithms that allow for deeper comprehension, and reduced biases when compared to its predecessors. Llama 3.3 excels in various domains such as natural language understanding, creative writing, technical writing, and multilingual conversations, making it an invaluable tool for businesses, developers, and researchers. Furthermore, its modular design lends itself to adaptable deployment across specific sectors, ensuring consistent performance and flexibility even in expansive applications. With these significant improvements, Llama 3.3 is set to transform the benchmarks for AI language models and inspire further innovations in the field. It is an exciting time for AI development as this new version opens doors to novel possibilities in human-computer interaction.

LiteLLM acts as an all-encompassing platform that streamlines interaction with over 100 Large Language Models (LLMs) through a unified interface. It features a Proxy Server (LLM Gateway) alongside a Python SDK, empowering developers to seamlessly integrate various LLMs into their applications. The Proxy Server adopts a centralized management system that facilitates load balancing, cost monitoring across multiple projects, and guarantees alignment of input/output formats with OpenAI standards. By supporting a diverse array of providers, it enhances operational management through the creation of unique call IDs for each request, which is vital for effective tracking and logging in different systems. Furthermore, developers can take advantage of pre-configured callbacks to log data using various tools, which significantly boosts functionality. For enterprise users, LiteLLM offers an array of advanced features such as Single Sign-On (SSO), extensive user management capabilities, and dedicated support through platforms like Discord and Slack, ensuring businesses have the necessary resources for success. This comprehensive strategy not only heightens operational efficiency but also cultivates a collaborative atmosphere where creativity and innovation can thrive, ultimately leading to better outcomes for all users. Thus, LiteLLM positions itself as a pivotal tool for organizations looking to leverage LLMs effectively in their workflows.

Meta’s Llama 4 Maverick is a state-of-the-art multimodal AI model that packs 17 billion active parameters and 128 experts into a high-performance solution. Its performance surpasses other top models, including GPT-4o and Gemini 2.0 Flash, particularly in reasoning, coding, and image processing benchmarks. Llama 4 Maverick excels at understanding and generating text while grounding its responses in visual data, making it perfect for applications that require both types of information. This model strikes a balance between power and efficiency, offering top-tier AI capabilities at a fraction of the parameter size compared to larger models, making it a versatile tool for developers and enterprises alike.

Llama 4 Scout represents a leap forward in multimodal AI, featuring 17 billion active parameters and a groundbreaking 10 million token context length. With its ability to integrate both text and image data, Llama 4 Scout excels at tasks like multi-document summarization, complex reasoning, and image grounding. It delivers superior performance across various benchmarks and is particularly effective in applications requiring both language and visual comprehension. Scout's efficiency and advanced capabilities make it an ideal solution for developers and businesses looking for a versatile and powerful model to enhance their AI-driven projects.

Qwen3, the latest large language model from the Qwen family, introduces a new level of flexibility and power for developers and researchers. With models ranging from the high-performance Qwen3-235B-A22B to the smaller Qwen3-4B, Qwen3 is engineered to excel across a variety of tasks, including coding, math, and natural language processing. The unique hybrid thinking modes allow users to switch between deep reasoning for complex tasks and fast, efficient responses for simpler ones. Additionally, Qwen3 supports 119 languages, making it ideal for global applications. The model has been trained on an unprecedented 36 trillion tokens and leverages cutting-edge reinforcement learning techniques to continually improve its capabilities. Available on multiple platforms, including Hugging Face and ModelScope, Qwen3 is an essential tool for those seeking advanced AI-powered solutions for their projects.

Nomic Embed is an extensive suite of open-source, high-performance embedding models designed for various applications, including multilingual text handling, multimodal content integration, and code analysis. Among these models, Nomic Embed Text v2 utilizes a Mixture-of-Experts (MoE) architecture that adeptly manages over 100 languages with an impressive 305 million active parameters, providing rapid inference capabilities. In contrast, Nomic Embed Text v1.5 offers adaptable embedding dimensions between 64 and 768 through Matryoshka Representation Learning, enabling developers to balance performance and storage needs effectively. For multimodal applications, Nomic Embed Vision v1.5 collaborates with its text models to form a unified latent space for both text and image data, significantly improving the ability to conduct seamless multimodal searches. Additionally, Nomic Embed Code demonstrates superior embedding efficiency across multiple programming languages, proving to be an essential asset for developers. This adaptable suite of models not only enhances workflow efficiency but also inspires developers to approach a wide range of challenges with creativity and innovation, thereby broadening the scope of what they can achieve in their projects.

BGE, or BAAI General Embedding, functions as a comprehensive toolkit designed to enhance search performance and support Retrieval-Augmented Generation (RAG) applications. It includes features for model inference, evaluation, and fine-tuning of both embedding models and rerankers, facilitating the development of advanced information retrieval systems. Among its key components are embedders and rerankers, which can seamlessly integrate into RAG workflows, leading to marked improvements in the relevance and accuracy of search outputs. BGE supports a range of retrieval strategies, such as dense retrieval, multi-vector retrieval, and sparse retrieval, which enables it to adjust to various data types and retrieval scenarios. Users can conveniently access these models through platforms like Hugging Face, and the toolkit provides an array of tutorials and APIs for efficient implementation and customization of retrieval systems. By leveraging BGE, developers can create resilient and high-performance search solutions tailored to their specific needs, ultimately enhancing the overall user experience and satisfaction. Additionally, the inherent flexibility of BGE guarantees its capability to adapt to new technologies and methodologies as they emerge within the data retrieval field, ensuring its continued relevance and effectiveness. This adaptability not only meets current demands but also anticipates future trends in information retrieval.

ZenCtrl, developed by Fotographer AI, is a groundbreaking open-source toolkit designed for AI image generation, enabling the creation of high-quality visuals from a single input image without necessitating any prior training. This innovative tool facilitates accurate regeneration of objects and subjects from multiple viewpoints and backgrounds, providing real-time element regeneration that enhances both stability and flexibility during the creative process. Users can effortlessly regenerate subjects from various angles, swap backgrounds or outfits with just a click, and begin producing results immediately, bypassing the need for extensive training. Leveraging advanced image processing techniques, ZenCtrl ensures high precision while reducing the dependency on large training datasets. Its architecture comprises streamlined sub-models, each finely tuned for specific tasks, leading to a lightweight system that yields sharper and more controllable results. The latest version of ZenCtrl brings substantial enhancements to the generation of both subjects and backgrounds, guaranteeing that the final images are not only coherent but also visually captivating. This ongoing improvement demonstrates a dedication to equipping users with the most effective and efficient tools for their creative projects, ensuring that they can achieve their desired outcomes with ease. As the toolkit evolves, users can expect even more features and capabilities that will further streamline their creative workflows.

In recent times, we have been genuinely appreciative of the substantial feedback received, and we are committed to executing a launch that prioritizes responsibility and security, taking into account the valuable insights acquired from beta testing and community input for our developers to integrate. By working hand in hand with the dedicated legal, ethics, and technology teams at HuggingFace, alongside the talented engineers at CoreWeave, we have successfully developed an integrated AI Safety Classifier within our software package. This classifier is specifically engineered to understand diverse concepts and factors during content generation, allowing it to screen outputs that may not meet user expectations. Users have the flexibility to modify the parameters of this feature, and we wholeheartedly welcome suggestions from the community for further improvements. Although image generation models exhibit remarkable potential, there is still an ongoing necessity for progress in accurately aligning results with our desired objectives. Our ultimate aim remains to enhance these tools continually, ensuring they effectively adapt to the changing requirements of users and foster a collaborative environment for innovation.

Canopy Labs has introduced Orpheus, a groundbreaking collection of advanced speech large language models (LLMs) designed to replicate human-like speech generation. Built on the Llama-3 architecture, these models have been developed using a vast dataset of over 100,000 hours of English speech, enabling them to produce output with natural intonation, emotional nuance, and a rhythmic quality that surpasses current high-end closed-source models. One of the standout features of Orpheus is its zero-shot voice cloning capability, which allows users to replicate voices without needing any prior fine-tuning, alongside user-friendly tags that assist in manipulating emotion and intonation. Engineered for minimal latency, these models achieve around 200ms streaming latency for real-time applications, with potential reductions to approximately 100ms when input streaming is employed. Canopy Labs offers both pre-trained and fine-tuned models featuring 3 billion parameters under the adaptable Apache 2.0 license, and there are plans to develop smaller models with 1 billion, 400 million, and 150 million parameters to accommodate devices with limited processing power. This initiative is anticipated to enhance accessibility and expand the range of applications across diverse platforms and scenarios, making advanced speech generation technology more widely available. As technology continues to evolve, the implications of such advancements could significantly influence fields such as entertainment, education, and customer service.

CAMB.AI's MARS6 marks a groundbreaking leap in text-to-speech (TTS) technology, emerging as the first speech model accessible on the Amazon Web Services (AWS) Bedrock platform. This integration enables developers to seamlessly incorporate advanced TTS features into their generative AI projects, opening avenues for more engaging voice assistants, enthralling audiobooks, interactive media, and a range of audio-centric experiences. Leveraging innovative algorithms, MARS6 produces speech synthesis that is both natural and expressive, setting a new standard for TTS quality. Developers can easily utilize MARS6 through the Amazon Bedrock platform, which facilitates smooth integration into their applications, thus improving user engagement and making content more accessible. The introduction of MARS6 into the diverse collection of foundational models on AWS Bedrock underscores CAMB.AI's commitment to expanding the frontiers of machine learning and artificial intelligence. By equipping developers with the critical tools necessary for creating immersive audio experiences, CAMB.AI not only fosters innovation but also guarantees that these advancements are built on AWS's reliable and scalable infrastructure. This collaboration between cutting-edge TTS technology and cloud solutions is set to redefine user interaction with audio content across various platforms, enhancing the overall digital experience even further. With such transformative potential, MARS6 is positioned to lead the charge in the next generation of audio applications.

We are excited to announce the launch of Whisper, an open-source neural network that delivers accuracy and robustness in English speech recognition that rivals that of human abilities. This automatic speech recognition (ASR) system has been meticulously trained using a vast dataset of 680,000 hours of multilingual and multitask supervised data sourced from the internet. Our findings indicate that employing such a rich and diverse dataset greatly enhances the system's performance in adapting to various accents, background noise, and specialized jargon. Moreover, Whisper not only supports transcription in multiple languages but also offers translation capabilities into English from those languages. To facilitate the development of real-world applications and to encourage ongoing research in the domain of effective speech processing, we are providing access to both the models and the inference code. The Whisper architecture is designed with a simple end-to-end approach, leveraging an encoder-decoder Transformer framework. The input audio is segmented into 30-second intervals, which are then converted into log-Mel spectrograms before entering the encoder. By democratizing access to this technology, we aspire to inspire new advancements in the realm of speech recognition and its applications across different industries. Our commitment to open-source principles ensures that developers worldwide can collaboratively enhance and refine these tools for future innovations.

Stable Diffusion XL, commonly referred to as SDXL, is the latest iteration in image generation technology, purposefully crafted to deliver superior photorealism and intricate details in visual compositions compared to its predecessors, such as SD 2.1. This advancement empowers users to produce images with enhanced facial accuracy and more legible text, while also facilitating the generation of aesthetically pleasing artworks through brief prompts. Consequently, artists and creators are now able to articulate their concepts with greater clarity and efficiency, expanding the possibilities for creative expression in their work. The evolution of this model marks a significant milestone in the field of digital art generation, opening new avenues for innovation and creativity.

Mixedbread is a cutting-edge AI search engine designed to streamline the development of powerful AI search and Retrieval-Augmented Generation (RAG) applications for users. It provides a holistic AI search solution, encompassing vector storage, embedding and reranking models, as well as document parsing tools. By utilizing Mixedbread, users can easily transform unstructured data into intelligent search features that boost AI agents, chatbots, and knowledge management systems while keeping the process simple. The platform integrates smoothly with widely-used services like Google Drive, SharePoint, Notion, and Slack. Its vector storage capabilities enable users to set up operational search engines within minutes and accommodate a broad spectrum of over 100 languages. Mixedbread's embedding and reranking models have achieved over 50 million downloads, showcasing their exceptional performance compared to OpenAI in both semantic search and RAG applications, all while being open-source and cost-effective. Furthermore, the document parser adeptly extracts text, tables, and layouts from various formats like PDFs and images, producing clean, AI-ready content without the need for manual work. This efficiency and ease of use make Mixedbread the perfect solution for anyone aiming to leverage AI in their search applications, ensuring a seamless experience for users.