Top Mercury 2 Alternatives

Celeris-1 distinguishes itself as a rapid and adaptable language model platform, enhanced by a diffusion model that provides state-of-the-art intelligence at remarkable speeds. In contrast to traditional autoregressive models that produce tokens one after another, Celeris utilizes a diffusion-based inference architecture that facilitates concurrent generation, leading to response times that can be recorded in just milliseconds. On the MMLU-Pro benchmark, Celeris-1 achieves an impressive accuracy rate of 75.9%, with a median response time of 158 milliseconds and an extraordinary output rate of 1,664 tokens per second, placing it in close proximity to top models while functioning more than ten times faster. This robust model is available through an API compatible with OpenAI, making it easy for developers to integrate it into their existing SDKs and applications with minimal effort. Moreover, it features streaming capabilities that cater to real-time applications, enabling response times as quick as 24 milliseconds without any interruptions or delays, which makes it particularly suitable for interactive scenarios. Additionally, Celeris-1’s innovative architecture not only enhances its performance but also sets a new standard for future language model development. Overall, Celeris-1 signifies a remarkable leap forward in the efficiency and capability of language models.

The Nemotron 3 Nano, a compact yet robust language model from NVIDIA's Nemotron 3 lineup, is specifically designed to excel in agentic reasoning, engaging dialogue, and programming tasks. Its cutting-edge Mixture-of-Experts Mamba-Transformer architecture selectively activates a specific subset of parameters for each token, allowing for quick inference times while maintaining high accuracy and reasoning skills. With an impressive total of around 31.6 billion parameters, including about 3.2 billion active ones (or 3.6 billion when including embeddings), this model outperforms its predecessor, the Nemotron 2 Nano, while demanding less computational power for every forward pass. It boasts the capability to handle long-context processing of up to one million tokens, enabling it to efficiently analyze lengthy documents, navigate complex workflows, and carry out detailed reasoning tasks in one go. Additionally, it is designed for high-throughput, real-time performance, making it particularly skilled in managing multi-turn dialogues, executing tool invocations, and handling agent-driven workflows that require sophisticated planning and reasoning. This adaptability renders the Nemotron 3 Nano a top-tier option for a wide range of applications that necessitate advanced cognitive functions and seamless interaction. Its ability to integrate these features sets a new standard in the landscape of language models.

Mercury Edit 2 is an advanced AI model developed by Inception Labs, forming part of the Mercury suite, and is designed for efficient reasoning, coding, and editing through a unique architecture that diverges from standard large language models. This model improves upon the capabilities of Mercury 2, a diffusion-based system that can produce and enhance entire outputs at once, as opposed to the traditional approach of generating text token by token, resulting in significantly faster processing and more flexible editing. Rather than serving as a straightforward "typewriter," it functions as a responsive editor, starting with an initial draft and progressively refining it across multiple tokens in tandem, which allows for immediate interaction and rapid iterations in various areas, including code refinement, content generation, and agent-oriented tasks. With a remarkable throughput of nearly 1,000 tokens per second, this framework greatly exceeds the performance of conventional models while maintaining strong reasoning capabilities across a variety of benchmarks. Its innovative structure not only changes how users engage with AI but also establishes a new benchmark for excellence within the realm of artificial intelligence, pushing the boundaries of what is possible in this rapidly evolving field. As a result, it opens up new avenues for creativity and productivity that were previously unattainable.

Mercury, an innovative development from Inception Labs, is the first large language model designed for commercial use that harnesses diffusion technology, achieving an impressive tenfold enhancement in processing speed while simultaneously reducing costs when compared to traditional autoregressive models. Built for outstanding capabilities in reasoning, coding, and structured text generation, Mercury can process over 1000 tokens per second on NVIDIA H100 GPUs, making it one of the fastest models available today. Unlike conventional models that generate text in a sequential manner, Mercury employs a coarse-to-fine diffusion strategy to refine its outputs, which not only increases accuracy but also reduces the frequency of hallucinations. Furthermore, the introduction of Mercury Coder, a specialized coding module, allows developers to leverage cutting-edge AI-assisted code generation that is both swift and efficient. This pioneering methodology not only revolutionizes coding techniques but also establishes a new standard for what AI can achieve across diverse applications, showcasing its versatility and potential. As a result, Mercury is positioned to lead the evolution of AI technology in various fields, promising to enhance productivity and innovation significantly.

Seed Diffusion Preview represents a cutting-edge language model tailored for code generation that utilizes discrete-state diffusion, enabling it to generate code in a non-linear fashion, which significantly accelerates inference times without sacrificing quality. This pioneering methodology follows a two-phase training procedure that consists of mask-based corruption coupled with edit-based enhancement, allowing a typical dense Transformer to strike an optimal balance between efficiency and accuracy while steering clear of shortcuts such as carry-over unmasking, thereby ensuring rigorous density estimation. Remarkably, the model achieves an impressive inference rate of 2,146 tokens per second on H20 GPUs, outperforming existing diffusion benchmarks while either matching or exceeding accuracy on recognized code evaluation metrics, including various editing tasks. This exceptional performance not only establishes a new standard for the trade-off between speed and quality in code generation but also highlights the practical effectiveness of discrete diffusion techniques in real-world coding environments. Furthermore, its achievements pave the way for improved productivity in coding tasks across diverse platforms, potentially transforming how developers approach code generation and refinement.

DiffusionGemma is a groundbreaking open model that delves into the phenomenon of text diffusion, offering an exceptionally quick approach to text generation. Licensed under Apache 2.0, this model features a staggering 26 billion parameters and utilizes a Mixture of Experts (MoE) architecture, pushing the boundaries beyond the conventional sequential token generation found in autoregressive models. Rather than generating tokens one by one, it is capable of producing complete blocks of text simultaneously, yielding generation speeds that can be up to four times quicker on GPUs. With foundations rooted in the parameter efficiency of the Gemma 4 family and insights from Gemini Diffusion research, DiffusionGemma boasts a distinctive diffusion head that significantly accelerates the generation process. Its design targets researchers and developers focused on optimizing local workflows that demand speed, such as in-line editing, rapid iterations, and complex narrative structures. By shifting the decoding bottleneck from memory bandwidth to computational capacity, the model can generate over 1,000 tokens per second on a single NVIDIA H100 and more than 700 tokens per second when utilizing an NVIDIA GeForce RTX 5090. This advancement not only enhances efficiency in text generation but also opens up new possibilities for various applications in the realm of natural language processing, paving the way for innovative developments in the field. Ultimately, the capabilities of DiffusionGemma could lead to transformative changes in how we approach text generation tasks.

GLM-Image is a cutting-edge, open-source image generation model developed by Z.ai that seamlessly integrates deep linguistic understanding with exceptional visual output. Unlike traditional diffusion models, it utilizes a unique hybrid approach that combines an autoregressive language model with a diffusion decoder, enabling it to thoroughly analyze the structure, semantics, and relationships within a given prompt prior to generating the respective image. This innovative design makes GLM-Image especially proficient in scenarios that require precise semantic control, such as the development of infographics, presentation materials, posters, and diagrams that incorporate detailed text and complex layouts. Featuring around 16 billion parameters, the model excels in producing clear, well-placed text within images—an area where many competitors struggle—while maintaining high visual quality and coherence. This remarkable blend of features establishes GLM-Image as an indispensable resource for professionals aiming to craft visually striking and textually rich content. Ultimately, its sophisticated capabilities and user-friendly interface make it an attractive option for a variety of creative projects.

Luma AI has introduced UNI-1, a revolutionary multimodal AI model that integrates visual generation and reasoning into a single framework, representing a significant step toward achieving multimodal general intelligence. This pioneering structure tackles the limitations faced by traditional AI systems, where distinct components such as language models and image generators operate separately, resulting in a lack of cohesive reasoning. By fusing these capabilities, UNI-1 promotes fluid interaction among language understanding, visual interpretation, and image production, enabling the model to logically analyze scenes, execute commands, and generate visuals that conform to both logical and spatial requirements. At the core of this system is a decoder-only autoregressive transformer that manages both text and images as an integrated sequence of tokens, which allows for a harmonious interaction between linguistic and visual information. This innovative integration not only boosts the efficiency of the AI model but also expands its potential applications across a wide range of fields, paving the way for future advancements in artificial intelligence. Ultimately, UNI-1 redefines the possibilities of multimodal AI, bringing us closer to the realization of truly intelligent systems.

Inception Labs is pioneering the evolution of artificial intelligence with its cutting-edge development of diffusion-based large language models (dLLMs), which mark a major breakthrough in the industry by delivering performance that is up to ten times faster and costing five to ten times less than traditional autoregressive models. Inspired by the success of diffusion methods in creating images and videos, Inception's dLLMs provide enhanced reasoning capabilities, superior error correction, and the ability to handle multimodal inputs, all of which significantly improve the generation of structured and accurate text. This revolutionary methodology not only enhances efficiency but also increases user control over AI-generated content. Furthermore, with a diverse range of applications in business solutions, academic exploration, and content generation, Inception Labs is setting new standards for speed and effectiveness in AI-driven processes. These groundbreaking advancements hold the potential to transform numerous sectors by streamlining workflows and boosting overall productivity, ultimately leading to a more efficient future. As industries adapt to these innovations, the impact on operational dynamics is expected to be profound.

Gemini Diffusion embodies our innovative research effort focused on transforming the understanding of diffusion within language and text creation. Currently, large language models form the foundational technology behind generative AI. Through the application of a diffusion methodology, we are developing a novel language model that improves user agency, encourages creativity, and hastens the text generation process. In contrast to conventional models that generate text in a linear fashion, diffusion models utilize a distinctive method by producing results through the gradual refinement of noise. This iterative approach allows them to swiftly reach solutions and implement real-time adjustments during the generation phase. Consequently, they excel in various tasks, particularly in areas like editing, mathematics, and programming. Additionally, by generating complete token blocks simultaneously, they yield more cohesive responses to user inquiries than autoregressive models do. Notably, Gemini Diffusion's performance on external evaluations is competitive with that of significantly larger models, all while offering improved speed, marking it as a significant breakthrough in the domain. This advancement not only simplifies the generation process but also paves the way for new forms of creative expression in language-oriented applications, showcasing the potential of rethinking traditional methodologies.

The Nemotron-3 Super stands out as a groundbreaking addition to NVIDIA's Nemotron 3 series of open models, designed specifically to support advanced agentic AI systems capable of reasoning, planning, and executing complex multi-step workflows in challenging settings. It incorporates a distinctive hybrid Mamba-Transformer Mixture-of-Experts architecture that combines the streamlined capabilities of Mamba layers with the contextual richness offered by transformer attention mechanisms, enabling it to effectively handle long sequences and complicated reasoning tasks with notable precision and efficiency. By activating only a selected subset of its parameters for each token, this design greatly improves computational efficiency while ensuring strong reasoning skills, making it particularly suitable for scalable inference in demanding situations. With an impressive configuration of around 120 billion parameters, of which approximately 12 billion are engaged during inference, the Nemotron-3 Super significantly enhances its capacity for managing multi-step reasoning and facilitating collaborative interactions among agents in broad contexts. This combination of features not only empowers it to address a wide array of challenges in the AI landscape but also positions it as a key player in the evolution of intelligent systems. Overall, the model exemplifies the potential for future innovations in AI technology.

The Phi-4-mini-flash-reasoning model, boasting 3.8 billion parameters, is a key part of Microsoft's Phi series, tailored for environments with limited processing capabilities such as edge and mobile platforms. Its state-of-the-art SambaY hybrid decoder architecture combines Gated Memory Units (GMUs) with Mamba state-space and sliding-window attention layers, resulting in performance improvements that are up to ten times faster and decreasing latency by two to three times compared to previous iterations, while still excelling in complex reasoning tasks. Designed to support a context length of 64K tokens and fine-tuned on high-quality synthetic datasets, this model is particularly effective for long-context retrieval and real-time inference, making it efficient enough to run on a single GPU. Accessible via platforms like Azure AI Foundry, NVIDIA API Catalog, and Hugging Face, Phi-4-mini-flash-reasoning presents developers with the tools to build applications that are both rapid and highly scalable, capable of performing intensive logical processing. This extensive availability encourages a diverse group of developers to utilize its advanced features, paving the way for creative and innovative application development in various fields.

On June 23, 2025, Microsoft introduced Mu, a cutting-edge language model boasting 330 million parameters and designed to significantly improve the agent experience in Windows environments by seamlessly converting natural language questions into functional calls for Settings, with all operations executed on-device via NPUs at an impressive speed exceeding 100 tokens per second while maintaining high accuracy. Utilizing Phi Silica optimizations, Mu's encoder-decoder architecture employs a fixed-length latent representation that notably minimizes computational requirements and memory consumption, achieving a 47 percent decrease in first-token latency and delivering a decoding speed that is 4.7 times faster on Qualcomm Hexagon NPUs in comparison to traditional decoder-only models. Furthermore, the model is enhanced by hardware-aware tuning methodologies, which incorporate a strategic 2/3–1/3 division of encoder and decoder parameters, shared weights for both input and output embeddings, Dual LayerNorm, rotary positional embeddings, and grouped-query attention, facilitating rapid inference rates that surpass 200 tokens per second on devices like the Surface Laptop 7, along with response times for settings-related queries that are under 500 ms. This impressive blend of features and optimizations establishes Mu as a revolutionary development in the realm of on-device language processing capabilities, setting new standards for speed and efficiency. As a result, users can expect a more intuitive and responsive experience when interacting with their Windows settings through natural language.

NVIDIA's Nemotron 3.5 Lightning represents an advanced mixture-of-experts model that features an impressive 30 billion parameters, with 3 billion of these actively engaged, and is specifically designed to deliver efficient, high-throughput performance for AI agents that operate continuously over extended periods. This model is crafted for the execution aspects of agentic systems, skillfully handling common tasks such as invoking tools, verifying outputs, carrying out routine commands, and assigning responsibilities to subagents, while larger reasoning models focus on strategic planning and orchestration. By utilizing a mixture-of-experts framework, it selectively engages a limited number of parameters for each input token, effectively combining the vast potential of a larger model with substantially decreased computational requirements. The training process is fine-tuned for popular agent harnesses, significantly improving inference speed through methods like speculative decoding, multi-token prediction, DFlash, and DSpark, which enhance its adaptability to various operational contexts. Moreover, it supports BF16 and NVFP4 checkpoints, ensuring deployment flexibility across platforms ranging from local systems such as DGX Spark and GeForce RTX hardware to large-scale data center environments. This innovative design not only amplifies AI capabilities but also positions Nemotron 3.5 Lightning as a pivotal resource for the evolution of intelligent systems, paving the way for future advancements in the field.

GPT-Realtime-2.1 is an OpenAI realtime reasoning model for developers building voice agents, conversational AI assistants, and speech-to-speech applications. The model is designed to support fast interactive experiences where users can speak naturally and receive audio or text responses. GPT-Realtime-2.1 updates GPT-Realtime-2 with improved handling of alphanumeric recognition, silence, background noise, and interruptions. It supports text, audio, and image input, with text and audio output, while video is not supported. The model includes configurable reasoning effort so developers can balance reasoning depth, latency, and token usage for different voice-agent workflows. GPT-Realtime-2.1 also supports instruction following, function calling, tool use, and reasoning tokens for more complex applications. Its 128,000-token context window and 32,000-token maximum output allow it to manage longer conversations and richer task context. OpenAI lists support across endpoints such as Chat Completions, Responses, Realtime, realtime translations, realtime transcription sessions, Assistants, Batch, and related API services. The model’s documented pricing includes $4 per 1 million text input tokens, $0.40 per 1 million cached text input tokens, and $24 per 1 million text output tokens, with separate audio and image token pricing. GPT-Realtime-2.1 is not documented as supporting streaming, structured outputs, fine-tuning, or predicted outputs. By combining realtime speech, multimodal input, reasoning, function calling, and tool use, GPT-Realtime-2.1 gives developers a foundation for building sophisticated AI voice agents and interactive customer-facing applications.

NVIDIA Cosmos is an innovative platform designed specifically for developers, featuring state-of-the-art generative World Foundation Models (WFMs), sophisticated video tokenizers, robust safety measures, and an efficient data processing and curation system that enhances the development of physical AI technologies. This platform equips developers engaged in fields like autonomous vehicles, robotics, and video analytics AI agents with the tools needed to generate highly realistic, physics-informed synthetic video data, drawing from a vast dataset that includes 20 million hours of both real and simulated footage. As a result, it allows for the quick simulation of future scenarios, the training of world models, and the customization of particular behaviors. The architecture of the platform consists of three main types of WFMs: Cosmos Predict, capable of generating up to 30 seconds of continuous video from diverse input modalities; Cosmos Transfer, which adapts simulations to function effectively across varying environments and lighting conditions, enhancing domain augmentation; and Cosmos Reason, a vision-language model that applies structured reasoning to interpret spatial-temporal data for effective planning and decision-making. Through these advanced capabilities, NVIDIA Cosmos not only accelerates the innovation cycle in physical AI applications but also promotes significant advancements across a wide range of industries, ultimately contributing to the evolution of intelligent technologies.

DeepSeek-V4-Flash is a next-generation Mixture-of-Experts language model engineered for high efficiency, scalability, and long-context intelligence. It consists of 284 billion total parameters with 13 billion activated parameters, enabling optimized performance with reduced computational overhead. The model supports an industry-leading context window of up to one million tokens, allowing it to process extensive datasets and complex workflows seamlessly. Its hybrid attention architecture combines advanced techniques to improve long-context efficiency and reduce memory usage. DeepSeek-V4-Flash is trained on over 32 trillion tokens, enhancing its capabilities in reasoning, coding, and knowledge-based tasks. It incorporates advanced optimization methods for stable training and faster convergence. The model supports multiple reasoning modes, including fast responses and deeper analytical processing for complex problems. While slightly less powerful than its Pro counterpart, it achieves comparable reasoning performance when given more computation budget. It is designed for agentic workflows, enabling multi-step reasoning and tool-based interactions. The model is well-suited for scalable deployments where performance and cost efficiency are both important. As an open-source solution, it offers flexibility for customization across various environments. It also reduces inference cost and resource usage compared to larger models. Overall, DeepSeek-V4-Flash delivers a strong balance of speed, efficiency, and capability for real-world AI use cases.

Step 3.5 Flash represents a state-of-the-art open-source foundational language model crafted for sophisticated reasoning and agent-like functionality, prioritizing efficiency; it employs a sparse Mixture of Experts (MoE) framework that activates roughly 11 billion of its nearly 196 billion parameters for each token, which ensures both dense intelligence and rapid responsiveness. The architecture includes a 3-way Multi-Token Prediction (MTP-3) system, enabling the generation of hundreds of tokens per second and supporting intricate multi-step reasoning and task execution, while efficiently handling extensive contexts through a hybrid sliding window attention technique that reduces computational stress on large datasets or codebases. Its remarkable capabilities in reasoning, coding, and agentic tasks often rival or exceed those of much larger proprietary models, further enhanced by a scalable reinforcement learning mechanism that promotes ongoing self-improvement. This innovative design not only highlights Step 3.5 Flash's effectiveness but also positions it as a transformative force in the domain of AI language models, indicating its vast potential across a plethora of applications. As such, it stands as a testament to the advancements in AI technology, paving the way for future developments.

Inkling-Small is an efficient multimodal AI model built to deliver strong reasoning and coding performance at a fraction of Inkling’s size. It is a Mixture-of-Experts transformer with 276 billion total parameters and 12 billion active parameters. The model was trained on NVIDIA GB300 NVL72 systems and is designed to combine high capability with more efficient inference. Inkling-Small supports native reasoning across text, images, and audio, allowing it to work across multimodal tasks without relying on separate encoders. Its context window supports up to one million tokens, making it useful for long-form reasoning, large-scale code understanding, document analysis, and agentic workflows. Users can adjust reasoning effort from minimal to extra high depending on whether they need faster responses or deeper computation. The model’s training process includes improved pre-training data, post-training with on-policy distillation from Inkling, and extended agentic coding reinforcement learning. These techniques helped Inkling-Small outperform its larger counterpart on reasoning and coding benchmarks. The model performs well in coding and tool-use harnesses and exceeds 80% on SWE-bench Verified. Its encoder-free architecture processes audio as dMel spectrograms and images as 40-by-40-pixel patches alongside text tokens. By combining efficient MoE design, one-million-token context, adjustable reasoning effort, multimodal processing, coding strength, and tool-use performance, Inkling-Small is designed for developers and teams that need capable AI with lower active compute requirements.

MiMo-V2-Flash is an advanced language model developed by Xiaomi that employs a Mixture-of-Experts (MoE) architecture, achieving a remarkable synergy between high performance and efficient inference. With an extensive 309 billion parameters, it activates only 15 billion during each inference, striking a balance between reasoning capabilities and computational efficiency. This model excels at processing lengthy contexts, making it particularly effective for tasks like long-document analysis, code generation, and complex workflows. Its unique hybrid attention mechanism combines sliding-window and global attention layers, which reduces memory usage while maintaining the capacity to grasp long-range dependencies. Moreover, the Multi-Token Prediction (MTP) feature significantly boosts inference speed by allowing multiple tokens to be processed in parallel. With the ability to generate around 150 tokens per second, MiMo-V2-Flash is specifically designed for scenarios requiring ongoing reasoning and multi-turn exchanges. The cutting-edge architecture of this model marks a noteworthy leap forward in language processing technology, demonstrating its potential applications across various domains. As such, it stands out as a formidable tool for developers and researchers alike.

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.

A streamlined model that excels in both text comprehension and multimodal reasoning abilities. The GPT-4o mini has been crafted to efficiently manage a vast range of tasks, characterized by its affordability and quick response times, which make it particularly suitable for scenarios requiring the simultaneous execution of multiple model calls, such as activating various APIs at once, analyzing large sets of information like complete codebases or lengthy conversation histories, and delivering prompt, real-time text interactions for customer support chatbots. At present, the API for GPT-4o mini supports both textual and visual inputs, with future enhancements planned to incorporate support for text, images, videos, and audio. This model features an impressive context window of 128K tokens and can produce outputs of up to 16K tokens per request, all while maintaining a knowledge base that is updated to October 2023. Furthermore, the advanced tokenizer utilized in GPT-4o enhances its efficiency in handling non-English text, thus expanding its applicability across a wider range of uses. Consequently, the GPT-4o mini is recognized as an adaptable resource for developers and enterprises, making it a valuable asset in various technological endeavors. Its flexibility and efficiency position it as a leader in the evolving landscape of AI-driven solutions.

Seed-Music is a comprehensive platform designed for the creation and modification of high-quality musical compositions, enabling users to produce both vocal and instrumental works from a variety of multimodal inputs, including lyrics, stylistic descriptions, sheet music, audio samples, or even vocal suggestions. This cutting-edge framework also supports the post-production editing of pre-existing tracks, allowing users to make direct modifications to melodies, instrumentations, timbres, or lyrics. It utilizes a combination of autoregressive language modeling and diffusion processes, structured into a three-phase pipeline: the first phase is representation learning, which encodes raw audio into intermediate formats such as audio tokens and symbolic music tokens; the second phase is generation, which converts these varied inputs into musical representations; and the final phase is rendering, which changes these representations into high-fidelity sound outputs. Additionally, Seed-Music's features encompass the transformation of lead sheets into complete songs, synthesis of singing voices, voice modulation, audio continuation, and style adaptation, offering users detailed control over the musical elements and composition. This extensive versatility positions it as an essential tool for musicians and music producers eager to delve into new realms of creativity and innovation. Ultimately, Seed-Music not only enhances the creative process but also broadens the possibilities for musical expression in the digital age.

Trinity Large Thinking is a cutting-edge open-source reasoning framework developed by Arcee AI, specifically designed for tackling complex, multi-step problems and workflows that involve autonomous agents requiring extensive planning and diverse tool utilization. With an impressive sparse Mixture-of-Experts architecture, it encompasses around 400 billion parameters, activating about 13 billion for each token, which not only boosts its operational efficiency but also fortifies its reasoning capabilities across various tasks, such as mathematical computations, code generation, and thorough analysis. A significant innovation of this model is its capacity for extended chain-of-thought reasoning, enabling it to generate intermediate "thinking traces" prior to presenting final results, which significantly enhances accuracy and dependability in intricate scenarios. Additionally, Trinity Large Thinking supports a generous context window of up to 262K tokens, which empowers it to effectively handle lengthy documents, maintain context during extended interactions, and operate smoothly within continuous agent loops. This exemplary design showcases a firm commitment to advancing the limits of automated reasoning systems, paving the way for more sophisticated applications in the future. As technology evolves, the potential for further enhancements in reasoning models like this one remains vast and exciting.

Qwen3.5 is an advanced open-weight multimodal AI system built to serve as the foundation for native digital agents capable of reasoning across text, images, and video. The primary release, Qwen3.5-397B-A17B, introduces a hybrid architecture that combines Gated DeltaNet linear attention with a sparse mixture-of-experts design, activating just 17 billion parameters per inference pass while maintaining a total parameter count of 397 billion. This selective activation dramatically improves decoding throughput and cost efficiency without sacrificing benchmark-level performance. Qwen3.5 demonstrates strong results across knowledge, multilingual reasoning, coding, STEM tasks, search agents, visual question answering, document understanding, and spatial intelligence benchmarks. The hosted Qwen3.5-Plus variant offers a default one-million-token context window and integrated tool usage such as web search and code interpretation for adaptive problem-solving. Expanded multilingual support now covers 201 languages and dialects, backed by a 250k vocabulary that enhances encoding and decoding efficiency across global use cases. The model is natively multimodal, using early fusion techniques and large-scale visual-text pretraining to outperform prior Qwen-VL systems in scientific reasoning and video analysis. Infrastructure innovations such as heterogeneous parallel training, FP8 precision pipelines, and disaggregated reinforcement learning frameworks enable near-text baseline throughput even with mixed multimodal inputs. Extensive reinforcement learning across diverse and generalized environments improves long-horizon planning, multi-turn interactions, and tool-augmented workflows. Designed for developers, researchers, and enterprises, Qwen3.5 supports scalable deployment through Alibaba Cloud Model Studio while paving the way toward persistent, economically aware, autonomous AI agents.

Mercury delivers robust subsystems intricately crafted to bring the latest innovations from Silicon Valley into the most challenging global environments. No matter the application, environment, or security requirements, Mercury's rugged servers and embedded processing solutions emerge as the sole dependable option for critical mission operations. The increasing necessity for compute-heavy applications, including AI, signals intelligence, and sensor fusion, has created a pressing demand for real-time big data processing at the network's edge. By transforming advanced technology from Silicon Valley into practical tools for the Aerospace and Defense (A&D) sector, Mercury's ruggedized servers and processing solutions facilitate actionable insights during real-time field operations. Each system is fully customizable, designed to expand the frontiers of computing innovation like never before. Furthermore, Mercury's airborne and mission computers not only boost the functionality of complex airborne applications but also streamline integration, technology upgrades, and safety certifications. With a strong emphasis on reliability and performance, Mercury guarantees that its technology remains trustworthy in the most challenging conditions. This commitment to excellence positions Mercury as a leader in providing high-performance solutions tailored to meet the evolving demands of the A&D industry.

Mercury serves as a sophisticated evolution of the BASIC programming language, offering complete compatibility with Microsoft Visual Basic.NET™ while broadening its scope and potential applications. This groundbreaking tool empowers developers to enhance their existing VB.NET projects, allowing them to leverage their Visual Basic™ skills for creating applications across a diverse range of contemporary platforms. Furthermore, developers have the option to integrate Mercury code with any of the other five Elements languages within a single project, enhancing functionality and versatility. The inclusion of the Mercury language in our development environments is both smooth and effective. You can utilize our smart yet powerful IDEs, Water for Windows or Fire for Mac, to create projects equipped with project templates, code completion features, and extensive debugging tools available on all platforms, among other advanced development options. Additionally, Mercury guarantees seamless compatibility with Visual Studio™ versions 2017, 2019, and 2022. Within the Elements framework, all programming languages are treated with equal importance, enabling easy integration of Mercury with C#, Swift, Java, Oxygene, and Go in the same project, which cultivates an atmosphere of flexibility and innovation in software creation. This adaptability not only enhances developer experience but also paves the way for exploring new and exciting programming paradigms and solutions.

VideoPoet is a groundbreaking modeling approach that enables any autoregressive language model or large language model (LLM) to function as a powerful video generator. This technique consists of several simple components. An autoregressive language model is trained to understand various modalities—including video, image, audio, and text—allowing it to predict the next video or audio token in a given sequence. The training structure for the LLM includes diverse multimodal generative learning objectives, which encompass tasks like text-to-video, text-to-image, image-to-video, video frame continuation, inpainting and outpainting of videos, video stylization, and video-to-audio conversion. Moreover, these tasks can be integrated to improve the model's zero-shot capabilities. This clear and effective methodology illustrates that language models can not only generate but also edit videos while maintaining impressive temporal coherence, highlighting their potential for sophisticated multimedia applications. Consequently, VideoPoet paves the way for a plethora of new opportunities in creative expression and automated content development, expanding the boundaries of how we produce and interact with digital media.

The NVIDIA Llama Nemotron family includes a range of advanced language models optimized for intricate reasoning tasks and a diverse set of agentic AI functions. These models excel in fields such as sophisticated scientific analysis, complex mathematics, programming, adhering to detailed instructions, and executing tool interactions. Engineered with flexibility in mind, they can be deployed across various environments, from data centers to personal computers, and they incorporate a feature that allows users to toggle reasoning capabilities, which reduces inference costs during simpler tasks. The Llama Nemotron series is tailored to address distinct deployment needs, building on the foundation of Llama models while benefiting from NVIDIA's advanced post-training methodologies. This results in a significant accuracy enhancement of up to 20% over the original models and enables inference speeds that can reach five times faster than other leading open reasoning alternatives. Such impressive efficiency not only allows for tackling more complex reasoning challenges but also enhances decision-making processes and substantially decreases operational costs for enterprises. Furthermore, the Llama Nemotron models stand as a pivotal leap forward in AI technology, making them ideal for organizations eager to incorporate state-of-the-art reasoning capabilities into their operations and strategies.

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.