Top MPT-7B Alternatives

Falcon-40B is a decoder-only model boasting 40 billion parameters, created by TII and trained on a massive dataset of 1 trillion tokens from RefinedWeb, along with other carefully chosen datasets. It is shared under the Apache 2.0 license, making it accessible for various uses. Why should you consider utilizing Falcon-40B? This model distinguishes itself as the premier open-source choice currently available, outpacing rivals such as LLaMA, StableLM, RedPajama, and MPT, as highlighted by its position on the OpenLLM Leaderboard. Its architecture is optimized for efficient inference and incorporates advanced features like FlashAttention and multiquery functionality, enhancing its performance. Additionally, the flexible Apache 2.0 license allows for commercial utilization without the burden of royalties or limitations. It's essential to recognize that this model is in its raw, pretrained state and is typically recommended to be fine-tuned to achieve the best results for most applications. For those seeking a version that excels in managing general instructions within a conversational context, Falcon-40B-Instruct might serve as a suitable alternative worth considering. Overall, Falcon-40B represents a formidable tool for developers looking to leverage cutting-edge AI technology in their projects.

Models designed to follow instructions, such as GPT-3.5 (text-DaVinci-003), ChatGPT, Claude, and Bing Chat, have experienced remarkable improvements in their functionalities, resulting in a notable increase in their utilization by users in various personal and professional environments. While their rising popularity and integration into everyday activities is evident, these models still face significant challenges, including the potential to spread misleading information, perpetuate detrimental stereotypes, and utilize offensive language. Addressing these pressing concerns necessitates active engagement from researchers and academics to further investigate these models. However, the pursuit of research on instruction-following models in academic circles has been complicated by the lack of accessible alternatives to proprietary systems like OpenAI’s text-DaVinci-003. To bridge this divide, we are excited to share our findings on Alpaca, an instruction-following language model that has been fine-tuned from Meta’s LLaMA 7B model, as we aim to enhance the dialogue and advancements in this domain. By shedding light on Alpaca, we hope to foster a deeper understanding of instruction-following models while providing researchers with a more attainable resource for their studies and explorations. This initiative marks a significant stride toward improving the overall landscape of instruction-following technologies.

We are excited to unveil the latest version of our open-source large language model, which includes model weights and initial code for the pretrained and fine-tuned Llama language models, ranging from 7 billion to 70 billion parameters. The Llama 2 pretrained models have been crafted using a remarkable 2 trillion tokens and boast double the context length compared to the first iteration, Llama 1. Additionally, the fine-tuned models have been refined through the insights gained from over 1 million human annotations. Llama 2 showcases outstanding performance compared to various other open-source language models across a wide array of external benchmarks, particularly excelling in reasoning, coding abilities, proficiency, and knowledge assessments. For its training, Llama 2 leveraged publicly available online data sources, while the fine-tuned variant, Llama-2-chat, integrates publicly accessible instruction datasets alongside the extensive human annotations mentioned earlier. Our project is backed by a robust coalition of global stakeholders who are passionate about our open approach to AI, including companies that have offered valuable early feedback and are eager to collaborate with us on Llama 2. The enthusiasm surrounding Llama 2 not only highlights its advancements but also marks a significant transformation in the collaborative development and application of AI technologies. This collective effort underscores the potential for innovation that can emerge when the community comes together to share resources and insights.

Dolly stands out as a cost-effective large language model, showcasing an impressive capability for following instructions akin to that of ChatGPT. The research conducted by the Alpaca team has shown that advanced models can be trained to significantly improve their adherence to high-quality instructions; however, our research suggests that even earlier open-source models can exhibit exceptional behavior when fine-tuned with a limited amount of instructional data. By making slight modifications to an existing open-source model containing 6 billion parameters from EleutherAI, Dolly has been enhanced to better follow instructions, demonstrating skills such as brainstorming and text generation that were previously lacking. This strategy not only emphasizes the untapped potential of older models but also invites exploration into new and innovative uses of established technologies. Furthermore, the success of Dolly encourages further investigation into how legacy models can be repurposed to meet contemporary needs effectively.

OpenLLaMA is a freely available version of Meta AI's LLaMA 7B, crafted using the RedPajama dataset. The model weights provided can easily substitute the LLaMA 7B in existing applications. Furthermore, we have also developed a streamlined 3B variant of the LLaMA model, catering to users who prefer a more compact option. This initiative enhances user flexibility by allowing them to select the most suitable model according to their particular requirements, thus accommodating a wider range of applications and use cases.

Foundation models, such as GPT-4, have propelled the field of artificial intelligence forward at an unprecedented pace; however, the most sophisticated models continue to be either restricted or only partially available to the public. To counteract this issue, the RedPajama initiative is focused on creating a suite of high-quality, completely open-source models. We are excited to share that we have successfully finished the first stage of this project: the recreation of the LLaMA training dataset, which encompasses over 1.2 trillion tokens. At present, a significant portion of leading foundation models is confined within commercial APIs, which limits opportunities for research and customization, especially when dealing with sensitive data. The pursuit of fully open-source models may offer a viable remedy to these constraints, on the condition that the open-source community can enhance the quality of these models to compete with their closed counterparts. Recent developments have indicated that there is encouraging progress in this domain, hinting that the AI sector may be on the brink of a revolutionary shift similar to what was seen with the introduction of Linux. The success of Stable Diffusion highlights that open-source alternatives can not only compete with high-end commercial products like DALL-E but also foster extraordinary creativity through the collaborative input of various communities. By nurturing a thriving open-source ecosystem, we can pave the way for new avenues of innovation and ensure that access to state-of-the-art AI technology is more widely available, ultimately democratizing the capabilities of artificial intelligence for all users.

StarCoder and StarCoderBase are sophisticated Large Language Models crafted for coding tasks, built from freely available data sourced from GitHub, which includes an extensive array of over 80 programming languages, along with Git commits, GitHub issues, and Jupyter notebooks. Similarly to LLaMA, these models were developed with around 15 billion parameters trained on an astonishing 1 trillion tokens. Additionally, StarCoderBase was specifically optimized with 35 billion Python tokens, culminating in the evolution of what we now recognize as StarCoder. Our assessments revealed that StarCoderBase outperforms other open-source Code LLMs when evaluated against well-known programming benchmarks, matching or even exceeding the performance of proprietary models like OpenAI's code-cushman-001 and the original Codex, which was instrumental in the early development of GitHub Copilot. With a remarkable context length surpassing 8,000 tokens, the StarCoder models can manage more data than any other open LLM available, thus unlocking a plethora of possibilities for innovative applications. This adaptability is further showcased by our ability to engage with the StarCoder models through a series of interactive dialogues, effectively transforming them into versatile technical aides capable of assisting with a wide range of programming challenges. Furthermore, this interactive capability enhances user experience, making it easier for developers to obtain immediate support and insights on complex coding issues.

Mistral 7B is a cutting-edge language model boasting 7.3 billion parameters, which excels in various benchmarks, even surpassing larger models such as Llama 2 13B. It employs advanced methods like Grouped-Query Attention (GQA) to enhance inference speed and Sliding Window Attention (SWA) to effectively handle extensive sequences. Available under the Apache 2.0 license, Mistral 7B can be deployed across multiple platforms, including local infrastructures and major cloud services. Additionally, a unique variant called Mistral 7B Instruct has demonstrated exceptional abilities in task execution, consistently outperforming rivals like Llama 2 13B Chat in certain applications. This adaptability and performance make Mistral 7B a compelling choice for both developers and researchers seeking efficient solutions. Its innovative features and strong results highlight the model's potential impact on natural language processing projects.

Sky-T1-32B-Preview represents a groundbreaking open-source reasoning model developed by the NovaSky team at UC Berkeley's Sky Computing Lab. It achieves performance levels similar to those of proprietary models like o1-preview across a range of reasoning and coding tests, all while being created for under $450, emphasizing its potential to provide advanced reasoning skills at a lower cost. Fine-tuned from Qwen2.5-32B-Instruct, this model was trained on a carefully selected dataset of 17,000 examples that cover diverse areas, including mathematics and programming. The training was efficiently completed in a mere 19 hours with the aid of eight H100 GPUs using DeepSpeed Zero-3 offloading technology. Notably, every aspect of this project—spanning data, code, and model weights—is fully open-source, enabling both the academic and open-source communities to not only replicate but also enhance the model's functionalities. Such openness promotes a spirit of collaboration and innovation within the artificial intelligence research and development landscape, inviting contributions from various sectors. Ultimately, this initiative represents a significant step forward in making powerful AI tools more accessible to a wider audience.

This repository presents the research preview for LongLLaMA, an innovative large language model capable of handling extensive contexts, reaching up to 256,000 tokens or potentially even more. Built on the OpenLLaMA framework, LongLLaMA has been fine-tuned using the Focused Transformer (FoT) methodology. The foundational code for this model comes from Code Llama. We are excited to introduce a smaller 3B base version of the LongLLaMA model, which is not instruction-tuned, and it will be released under an open license (Apache 2.0). Accompanying this release is inference code that supports longer contexts, available on Hugging Face. The model's weights are designed to effortlessly integrate with existing systems tailored for shorter contexts, particularly those that accommodate up to 2048 tokens. In addition to these features, we provide evaluation results and comparisons to the original OpenLLaMA models, thus offering a thorough insight into LongLLaMA's effectiveness in managing long-context tasks. This advancement marks a significant step forward in the field of language models, enabling more sophisticated applications and research opportunities.

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.

Kimi K2 showcases a groundbreaking series of open-source large language models that employ a mixture-of-experts (MoE) architecture, featuring an impressive total of 1 trillion parameters, with 32 billion parameters activated specifically for enhanced task performance. With the Muon optimizer at its core, this model has been trained on an extensive dataset exceeding 15.5 trillion tokens, and its capabilities are further amplified by MuonClip’s attention-logit clamping mechanism, enabling outstanding performance in advanced knowledge comprehension, logical reasoning, mathematics, programming, and various agentic tasks. Moonshot AI offers two unique configurations: Kimi-K2-Base, which is tailored for research-level fine-tuning, and Kimi-K2-Instruct, designed for immediate use in chat and tool interactions, thus allowing for both customized development and the smooth integration of agentic functionalities. Comparative evaluations reveal that Kimi K2 outperforms many leading open-source models and competes strongly against top proprietary systems, particularly in coding tasks and complex analysis. Additionally, it features an impressive context length of 128 K tokens, compatibility with tool-calling APIs, and support for widely used inference engines, making it a flexible solution for a range of applications. The innovative architecture and features of Kimi K2 not only position it as a notable achievement in artificial intelligence language processing but also as a transformative tool that could redefine the landscape of how language models are utilized in various domains. This advancement indicates a promising future for AI applications, suggesting that Kimi K2 may lead the way in setting new standards for performance and versatility in the industry.

OLMo 2 is a suite of fully open language models developed by the Allen Institute for AI (AI2), designed to provide researchers and developers with straightforward access to training datasets, open-source code, reproducible training methods, and extensive evaluations. These models are trained on a remarkable dataset consisting of up to 5 trillion tokens and are competitive with leading open-weight models such as Llama 3.1, especially in English academic assessments. A significant emphasis of OLMo 2 lies in maintaining training stability, utilizing techniques to reduce loss spikes during prolonged training sessions, and implementing staged training interventions to address capability weaknesses in the later phases of pretraining. Furthermore, the models incorporate advanced post-training methodologies inspired by AI2's Tülu 3, resulting in the creation of OLMo 2-Instruct models. To support continuous enhancements during the development lifecycle, an actionable evaluation framework called the Open Language Modeling Evaluation System (OLMES) has been established, featuring 20 benchmarks that assess vital capabilities. This thorough methodology not only promotes transparency but also actively encourages improvements in the performance of language models, ensuring they remain at the forefront of AI advancements. Ultimately, OLMo 2 aims to empower the research community by providing resources that foster innovation and collaboration in language modeling.

Code Llama is a sophisticated language model engineered to produce code from text prompts, setting itself apart as a premier choice among publicly available models for coding applications. This groundbreaking model not only enhances productivity for seasoned developers but also supports newcomers in tackling the complexities of learning programming. Its adaptability allows Code Llama to serve as both an effective productivity tool and a pedagogical resource, enabling programmers to develop more efficient and well-documented software. Furthermore, users can generate code alongside natural language explanations by inputting either format, which contributes to its flexibility for various programming tasks. Offered for free for both research and commercial use, Code Llama is based on the Llama 2 architecture and is available in three specific versions: the core Code Llama model, Code Llama - Python designed exclusively for Python development, and Code Llama - Instruct, which is fine-tuned to understand and execute natural language commands accurately. As a result, Code Llama stands out not just for its technical capabilities but also for its accessibility and relevance to diverse coding scenarios.

We are excited to unveil Mistral NeMo, our latest and most sophisticated small model, boasting an impressive 12 billion parameters and a vast context length of 128,000 tokens, all available under the Apache 2.0 license. In collaboration with NVIDIA, Mistral NeMo stands out in its category for its exceptional reasoning capabilities, extensive world knowledge, and coding skills. Its architecture adheres to established industry standards, ensuring it is user-friendly and serves as a smooth transition for those currently using Mistral 7B. To encourage adoption by researchers and businesses alike, we are providing both pre-trained base models and instruction-tuned checkpoints, all under the Apache license. A remarkable feature of Mistral NeMo is its quantization awareness, which enables FP8 inference while maintaining high performance levels. Additionally, the model is well-suited for a range of global applications, showcasing its ability in function calling and offering a significant context window. When benchmarked against Mistral 7B, Mistral NeMo demonstrates a marked improvement in comprehending and executing intricate instructions, highlighting its advanced reasoning abilities and capacity to handle complex multi-turn dialogues. Furthermore, its design not only enhances its performance but also positions it as a formidable option for multi-lingual tasks, ensuring it meets the diverse needs of various use cases while paving the way for future innovations.

Reka Flash 3 stands as a state-of-the-art multimodal AI model, boasting 21 billion parameters and developed by Reka AI, to excel in diverse tasks such as engaging in general conversations, coding, adhering to instructions, and executing various functions. This innovative model skillfully processes and interprets a wide range of inputs, which includes text, images, video, and audio, making it a compact yet versatile solution fit for numerous applications. Constructed from the ground up, Reka Flash 3 was trained on a diverse collection of datasets that include both publicly accessible and synthetic data, undergoing a thorough instruction tuning process with carefully selected high-quality information to refine its performance. The concluding stage of its training leveraged reinforcement learning techniques, specifically the REINFORCE Leave One-Out (RLOO) method, which integrated both model-driven and rule-oriented rewards to enhance its reasoning capabilities significantly. With a remarkable context length of 32,000 tokens, Reka Flash 3 effectively competes against proprietary models such as OpenAI's o1-mini, making it highly suitable for applications that demand low latency or on-device processing. Operating at full precision, the model requires a memory footprint of 39GB (fp16), but this can be optimized down to just 11GB through 4-bit quantization, showcasing its flexibility across various deployment environments. Furthermore, Reka Flash 3's advanced features ensure that it can adapt to a wide array of user requirements, thereby reinforcing its position as a leader in the realm of multimodal AI technology. This advancement not only highlights the progress made in AI but also opens doors to new possibilities for innovation across different sectors.

CodeQwen acts as the programming equivalent of Qwen, a collection of large language models developed by the Qwen team at Alibaba Cloud. This model, which is based on a transformer architecture that operates purely as a decoder, has been rigorously pre-trained on an extensive dataset of code. It is known for its strong capabilities in code generation and has achieved remarkable results on various benchmarking assessments. CodeQwen can understand and generate long contexts of up to 64,000 tokens and supports 92 programming languages, excelling in tasks such as text-to-SQL queries and debugging operations. Interacting with CodeQwen is uncomplicated; users can start a dialogue with just a few lines of code leveraging transformers. The interaction is rooted in creating the tokenizer and model using pre-existing methods, utilizing the generate function to foster communication through the chat template specified by the tokenizer. Adhering to our established guidelines, we adopt the ChatML template specifically designed for chat models. This model efficiently completes code snippets according to the prompts it receives, providing responses that require no additional formatting changes, thereby significantly enhancing the user experience. The smooth integration of these components highlights the adaptability and effectiveness of CodeQwen in addressing a wide range of programming challenges, making it an invaluable tool for developers.

Yi-Lightning, developed by 01.AI under the guidance of Kai-Fu Lee, represents a remarkable advancement in large language models, showcasing both superior performance and affordability. It can handle a context length of up to 16,000 tokens and boasts a competitive pricing strategy of $0.14 per million tokens for both inputs and outputs. This makes it an appealing option for a variety of users in the market. The model utilizes an enhanced Mixture-of-Experts (MoE) architecture, which incorporates meticulous expert segmentation and advanced routing techniques, significantly improving its training and inference capabilities. Yi-Lightning has excelled across diverse domains, earning top honors in areas such as Chinese language processing, mathematics, coding challenges, and complex prompts on chatbot platforms, where it achieved impressive rankings of 6th overall and 9th in style control. Its development entailed a thorough process of pre-training, focused fine-tuning, and reinforcement learning based on human feedback, which not only boosts its overall effectiveness but also emphasizes user safety. Moreover, the model features notable improvements in memory efficiency and inference speed, solidifying its status as a strong competitor in the landscape of large language models. This innovative approach sets the stage for future advancements in AI applications across various sectors.

Mistral Large 3 is a frontier-scale open AI model built on a sophisticated Mixture-of-Experts framework that unlocks 41B active parameters per step while maintaining a massive 675B total parameter capacity. This architecture lets the model deliver exceptional reasoning, multilingual mastery, and multimodal understanding at a fraction of the compute cost typically associated with models of this scale. Trained entirely from scratch on 3,000 NVIDIA H200 GPUs, it reaches competitive alignment performance with leading closed models, while achieving best-in-class results among permissively licensed alternatives. Mistral Large 3 includes base and instruction editions, supports images natively, and will soon introduce a reasoning-optimized version capable of even deeper thought chains. Its inference stack has been carefully co-designed with NVIDIA, enabling efficient low-precision execution, optimized MoE kernels, speculative decoding, and smooth long-context handling on Blackwell NVL72 systems and enterprise-grade clusters. Through collaborations with vLLM and Red Hat, developers gain an easy path to run Large 3 on single-node 8×A100 or 8×H100 environments with strong throughput and stability. The model is available across Mistral AI Studio, Amazon Bedrock, Azure Foundry, Hugging Face, Fireworks, OpenRouter, Modal, and more, ensuring turnkey access for development teams. Enterprises can go further with Mistral’s custom-training program, tailoring the model to proprietary data, regulatory workflows, or industry-specific tasks. From agentic applications to multilingual customer automation, creative workflows, edge deployment, and advanced tool-use systems, Mistral Large 3 adapts to a wide range of production scenarios. With this release, Mistral positions the 3-series as a complete family—spanning lightweight edge models to frontier-scale MoE intelligence—while remaining fully open, customizable, and performance-optimized across the stack.

DeepSeek-V2 represents an advanced Mixture-of-Experts (MoE) language model created by DeepSeek-AI, recognized for its economical training and superior inference efficiency. This model features a staggering 236 billion parameters, engaging only 21 billion for each token, and can manage a context length stretching up to 128K tokens. It employs sophisticated architectures like Multi-head Latent Attention (MLA) to enhance inference by reducing the Key-Value (KV) cache and utilizes DeepSeekMoE for cost-effective training through sparse computations. When compared to its earlier version, DeepSeek 67B, this model exhibits substantial advancements, boasting a 42.5% decrease in training costs, a 93.3% reduction in KV cache size, and a remarkable 5.76-fold increase in generation speed. With training based on an extensive dataset of 8.1 trillion tokens, DeepSeek-V2 showcases outstanding proficiency in language understanding, programming, and reasoning tasks, thereby establishing itself as a premier open-source model in the current landscape. Its groundbreaking methodology not only enhances performance but also sets unprecedented standards in the realm of artificial intelligence, inspiring future innovations in the field.

CodeGemma is an impressive collection of efficient and adaptable models that can handle a variety of coding tasks, such as middle code completion, code generation, natural language processing, mathematical reasoning, and instruction following. It includes three unique model variants: a 7B pre-trained model intended for code completion and generation using existing code snippets, a fine-tuned 7B version for converting natural language queries into code while following instructions, and a high-performing 2B pre-trained model that completes code at speeds up to twice as fast as its counterparts. Whether you are filling in lines, creating functions, or assembling complete code segments, CodeGemma is designed to assist you in any environment, whether local or utilizing Google Cloud services. With its training grounded in a vast dataset of 500 billion tokens, primarily in English and taken from web sources, mathematics, and programming languages, CodeGemma not only improves the syntactical precision of the code it generates but also guarantees its semantic accuracy, resulting in fewer errors and a more efficient debugging process. Beyond just functionality, this powerful tool consistently adapts and improves, making coding more accessible and streamlined for developers across the globe, thereby fostering a more innovative programming landscape. As the technology advances, users can expect even more enhancements in terms of speed and accuracy.

Olmo 3 constitutes an extensive series of open models that include versions with 7 billion and 32 billion parameters, delivering outstanding performance in areas such as base functionality, reasoning, instruction, and reinforcement learning, all while ensuring transparency throughout the development process, including access to raw training datasets, intermediate checkpoints, training scripts, extended context support (with a remarkable window of 65,536 tokens), and provenance tools. The backbone of these models is derived from the Dolma 3 dataset, which encompasses about 9 trillion tokens and employs a thoughtful mixture of web content, scientific research, programming code, and comprehensive documents; this meticulous strategy of pre-training, mid-training, and long-context usage results in base models that receive further refinement through supervised fine-tuning, preference optimization, and reinforcement learning with accountable rewards, leading to the emergence of the Think and Instruct versions. Importantly, the 32 billion Think model has earned recognition as the most formidable fully open reasoning model available thus far, showcasing a performance level that closely competes with that of proprietary models in disciplines such as mathematics, programming, and complex reasoning tasks, highlighting a considerable leap forward in the realm of open model innovation. This breakthrough not only emphasizes the capabilities of open-source models but also suggests a promising future where they can effectively rival conventional closed systems across a range of sophisticated applications, potentially reshaping the landscape of artificial intelligence.

LTM-2-mini is designed to manage a context of 100 million tokens, which is roughly equivalent to about 10 million lines of code or approximately 750 full-length novels. This model utilizes a sequence-dimension algorithm that proves to be around 1000 times more economical per decoded token compared to the attention mechanism employed by Llama 3.1 405B when operating within the same 100 million token context window. Additionally, the difference in memory requirements is even more pronounced; running Llama 3.1 405B with a 100 million token context requires an impressive 638 H100 GPUs per user just to sustain a single 100 million token key-value cache. In stark contrast, LTM-2-mini only needs a tiny fraction of the high-bandwidth memory available in one H100 GPU for the equivalent context, showcasing its remarkable efficiency. This significant advantage positions LTM-2-mini as an attractive choice for applications that require extensive context processing while minimizing resource usage. Moreover, the ability to efficiently handle such large contexts opens the door for innovative applications across various fields.

The Qwen2.5-1M language model, developed by the Qwen team, is an open-source innovation designed to handle extraordinarily long context lengths of up to one million tokens. This release features two model variations: Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M, marking a groundbreaking milestone as the first Qwen models optimized for such extensive token context. Moreover, the team has introduced an inference framework utilizing vLLM along with sparse attention mechanisms, which significantly boosts processing speeds for inputs of 1 million tokens, achieving speed enhancements ranging from three to seven times. Accompanying this model is a comprehensive technical report that delves into the design decisions and outcomes of various ablation studies. This thorough documentation ensures that users gain a deep understanding of the models' capabilities and the technology that powers them. Additionally, the improvements in processing efficiency are expected to open new avenues for applications needing extensive context management.

IBM® Granite™ offers a collection of AI models tailored for business use, developed with a strong emphasis on trustworthiness and scalability in AI solutions. At present, the open-source Granite models are readily available for use. Our mission is to democratize AI access for developers, which is why we have made the core Granite Code, along with Time Series, Language, and GeoSpatial models, available as open-source on Hugging Face. These resources are shared under the permissive Apache 2.0 license, enabling broad commercial usage without significant limitations. Each Granite model is crafted using carefully curated data, providing outstanding transparency about the origins of the training material. Furthermore, we have released tools for validating and maintaining the quality of this data to the public, adhering to the high standards necessary for enterprise applications. This unwavering commitment to transparency and quality not only underlines our dedication to innovation but also encourages collaboration within the AI community, paving the way for future advancements.

Amazon Nova Micro is a high-performance, text-only AI model that provides low-latency responses, making it ideal for applications needing real-time processing. With impressive capabilities in language understanding, translation, and reasoning, Nova Micro can generate over 200 tokens per second while maintaining high performance. This model supports fine-tuning on text inputs and is highly efficient, making it perfect for cost-conscious businesses looking to deploy AI for fast, interactive tasks such as code completion, brainstorming, and solving mathematical problems.

Teuken-7B is a cutting-edge multilingual language model designed to address the diverse linguistic landscape of Europe, emerging from the OpenGPT-X initiative. This model has been trained on a dataset where more than half comprises non-English content, effectively encompassing all 24 official languages of the European Union to ensure robust performance across these tongues. One of the standout features of Teuken-7B is its specially crafted multilingual tokenizer, which has been optimized for European languages, resulting in improved training efficiency and reduced inference costs compared to standard monolingual tokenizers. Users can choose between two distinct versions of the model: Teuken-7B-Base, which offers a foundational pre-trained experience, and Teuken-7B-Instruct, fine-tuned to enhance its responsiveness to user inquiries. Both variations are easily accessible on Hugging Face, promoting transparency and collaboration in the artificial intelligence sector while stimulating further advancements. The development of Teuken-7B not only showcases a commitment to fostering AI solutions but also underlines the importance of inclusivity and representation of Europe's rich cultural tapestry in technology. This initiative ultimately aims to bridge communication gaps and facilitate understanding among diverse populations across the continent.

GigaChat 3 Ultra is a breakthrough open-source LLM, offering 702 billion parameters built on an advanced MoE architecture that keeps computation efficient while delivering frontier-level performance. Its design activates only 36 billion parameters per step, combining high intelligence with practical deployment speeds, even for research and enterprise workloads. The model is trained entirely from scratch on a 14-trillion-token dataset spanning ten+ languages, expansive natural corpora, technical literature, competitive programming problems, academic datasets, and more than 5.5 trillion synthetic tokens engineered to enhance reasoning depth. This approach enables the model to achieve exceptional Russian-language capabilities, strong multilingual performance, and competitive global benchmark scores across math (GSM8K, MATH-500), programming (HumanEval+), and domain-specific evaluations. GigaChat 3 Ultra is optimized for compatibility with modern open-source tooling, enabling fine-tuning, inference, and integration using standard frameworks without complex custom builds. Advanced engineering techniques—including MTP, MLA, expert balancing, and large-scale distributed training—ensure stable learning at enormous scale while preserving fast inference. Beyond raw intelligence, the model includes upgraded alignment, improved conversational behavior, and a refined chat template using TypeScript-based function definitions for cleaner, more efficient interactions. It also features a built-in code interpreter, enhanced search subsystem with query reformulation, long-term user memory capabilities, and improved Russian-language stylistic accuracy down to punctuation and orthography. With leading performance on Russian benchmarks and strong showings across international tests, GigaChat 3 Ultra stands among the top five largest and most advanced open-source LLMs in the world. It represents a major engineering milestone for the open community.

Qwen2.5-Max is a cutting-edge Mixture-of-Experts (MoE) model developed by the Qwen team, trained on a vast dataset of over 20 trillion tokens and improved through techniques such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF). It outperforms models like DeepSeek V3 in various evaluations, excelling in benchmarks such as Arena-Hard, LiveBench, LiveCodeBench, and GPQA-Diamond, and also achieving impressive results in tests like MMLU-Pro. Users can access this model via an API on Alibaba Cloud, which facilitates easy integration into various applications, and they can also engage with it directly on Qwen Chat for a more interactive experience. Furthermore, Qwen2.5-Max's advanced features and high performance mark a remarkable step forward in the evolution of AI technology. It not only enhances productivity but also opens new avenues for innovation in the field.

GPT-5 mini is a faster, more affordable variant of OpenAI’s advanced GPT-5 language model, specifically tailored for well-defined and precise tasks that benefit from high reasoning ability. It accepts both text and image inputs (image input only), and generates high-quality text outputs, supported by a large 400,000-token context window and a maximum of 128,000 tokens in output, enabling complex multi-step reasoning and detailed responses. The model excels in providing rapid response times, making it ideal for use cases where speed and efficiency are critical, such as chatbots, customer service, or real-time analytics. GPT-5 mini’s pricing structure significantly reduces costs, with input tokens priced at $0.25 per million and output tokens at $2 per million, offering a more economical option compared to the flagship GPT-5. While it supports advanced features like streaming, function calling, structured output generation, and fine-tuning, it does not currently support audio input or image generation capabilities. GPT-5 mini integrates seamlessly with multiple API endpoints including chat completions, responses, embeddings, and batch processing, providing versatility for a wide array of applications. Rate limits are tier-based, scaling from 500 requests per minute up to 30,000 per minute for higher tiers, accommodating small to large scale deployments. The model also supports snapshots to lock in performance and behavior, ensuring consistency across applications. GPT-5 mini is ideal for developers and businesses seeking a cost-effective solution with high reasoning power and fast throughput. It balances cutting-edge AI capabilities with efficiency, making it a practical choice for applications demanding speed, precision, and scalability.