Top SubQ Snap Alternatives

MiniMax M3 is an open-weight multimodal foundation model from MiniMax that brings together coding capability, agentic reasoning, native multimodality, and long-context processing in one model. It is designed for demanding AI workflows where a system needs to understand large amounts of information, reason through multi-step tasks, use tools, and work with different input types. MiniMax M3 supports a context window of up to 1 million tokens, making it useful for large code repositories, long documents, multi-file analysis, research workflows, enterprise automation, and persistent agent memory. The model uses MiniMax Sparse Attention, an architecture built to improve efficiency at very long context lengths by reducing the cost of attention. MiniMax M3 is natively multimodal and can work with text, images, and video inputs, allowing it to support richer workflows than text-only language models. It is positioned for coding, software engineering, tool invocation, browser-style retrieval, computer-use-style tasks, and autonomous task decomposition. The model’s architecture includes a large total parameter count with a smaller number of activated parameters, supporting more efficient inference through a mixture-of-experts design. Developers can use MiniMax M3 to build coding assistants, AI agents, document intelligence systems, multimodal analysis tools, and automated enterprise workflows. Its long-context design helps reduce the need to compress or split large inputs, allowing teams to keep more project context available during reasoning. The model is available through open-weight releases and hosted API providers, giving developers multiple ways to test, deploy, or integrate it into applications. MiniMax M3 helps organizations build advanced AI systems that combine long memory, multimodal understanding, coding strength, and agentic execution.

Kimi K3 is Moonshot AI’s most advanced model, designed for high-end reasoning, software engineering, multimodal understanding, knowledge work, and agentic AI applications. The model has 2.8 trillion parameters and is built on Kimi Delta Attention, a hybrid linear attention mechanism created for long-context performance. It also uses Attention Residuals and supports a native context window of up to 1 million tokens. This makes Kimi K3 suitable for tasks involving large codebases, long research materials, enterprise documentation, multi-file analysis, legal documents, technical manuals, and complex workflows. Kimi K3 always has thinking mode enabled, with reasoning effort configured through the reasoning_effort field and maximum effort currently supported as the default. Developers can use the model through an OpenAI-compatible API, making it easier to integrate with existing SDKs, clients, and application infrastructure. The model supports streaming responses with separate reasoning and final-answer deltas, allowing applications to display reasoning progress and final content differently. Kimi K3 also supports strict structured output with JSON Schema, partial mode for continuing from a prefix, custom tool calling, required tool use, and dynamic tool loading through system messages. Its vision capabilities support image and video inputs through base64 or uploaded files, enabling analysis of visual content alongside text. Automatic context caching helps workflows that reuse long prefixes, such as large knowledge bases or persistent system context, without requiring developers to manage cache IDs manually. By combining frontier-scale parameters, long-context processing, visual input, structured outputs, tool orchestration, and developer-friendly API compatibility, Kimi K3 gives teams a strong foundation for advanced AI agents, coding assistants, research systems, enterprise automation, and multimodal applications.

DeepSeek-V4-Pro is a next-generation Mixture-of-Experts language model designed to deliver high performance across reasoning, coding, and long-context AI tasks. It features a massive architecture with 1.6 trillion total parameters and 49 billion activated parameters, enabling efficient computation while maintaining strong capabilities. The model supports an industry-leading context window of up to one million tokens, allowing it to process extremely large datasets, documents, and workflows. Its hybrid attention mechanism combines advanced techniques to optimize long-context efficiency and reduce computational requirements. DeepSeek-V4-Pro is trained on over 32 trillion tokens, enhancing its knowledge base and reasoning abilities. It incorporates advanced optimization methods to improve training stability and convergence. The model supports multiple reasoning modes, including fast responses and deep analytical thinking for complex problem solving. It performs strongly across benchmarks in coding, mathematics, and knowledge-based tasks. The architecture is designed for agentic workflows, enabling it to handle multi-step tasks and tool-based interactions. As an open-source model, it offers flexibility for customization and deployment across various environments. It also supports efficient memory usage and reduced inference costs compared to previous versions. The model’s capabilities make it suitable for both research and enterprise applications. Overall, DeepSeek-V4-Pro represents a significant advancement in scalable, high-performance AI with long-context intelligence.

GLM-5.3-Flash is an efficiency-focused multimodal AI model from Z.ai that combines advanced reasoning, coding, agentic execution, and visual intelligence. It is the first GLM-5-series model designed with native multimodal capabilities, allowing it to work directly with both textual and visual inputs. The architecture uses 320 billion total parameters while activating only 18 billion at a time, significantly reducing the amount of computation required for inference. A hybrid attention design blends linear attention for local information with sparse attention for retrieving important context from much larger inputs. Z.ai also uses technologies such as IndexPool and Manifold-Constrained Hyper-Connections to improve memory efficiency, latency, and model scaling. The model can operate with context windows of up to one million tokens, making it suitable for large repositories, lengthy documents, extended agent sessions, and complex multimodal workflows. GLM-5.3-Flash was trained on a 30-trillion-token multimodal corpus intended to strengthen reasoning across code, images, interfaces, documents, spreadsheets, presentations, and other business artifacts. In software development scenarios, the model can visually inspect rendered applications, evaluate its own output, and iteratively correct layout, functionality, or interaction issues. Z.ai’s reported benchmark results show large improvements over GLM-5.2 in areas such as software engineering and automation, while placing GLM-5.3-Flash close to leading frontier systems on several coding and agentic evaluations. Before its formal release, the model was anonymously tested under the name ox-alpha on OpenCode and OpenRouter, where Z.ai says it became one of the most widely used models during its testing period. GLM-5.3-Flash is available through Z.ai’s API and coding products as well as through downloadable weights on Hugging Face, with deployment support for SGLang, vLLM, and TokenSpeed.

Tev1 is an advanced model designed for quick decision-making and classification, created by Together AI and fine-tuned from Qwen3.5 4B. It features an intuitive interface that allows users to enter a state, pose a question, and select from 2 to 24 labeled options, leading to an output of a single letter that corresponds to a semantic key via software. This innovative structure is particularly advantageous for diverse production tasks, including the routing of support tickets, evaluating automated returns, implementing policy guidelines, conducting sentiment analysis, categorizing documents, and embedding precise decision-making within extensive agent workflows. With its impressive 4.7B parameter count and a context window of 32.8K tokens, Tev1 leverages non-thinking generation to deliver swift and dependable classifications. The model has been honed using a broad collection of 37,840 examples that encompass intent detection, boolean comprehension, sentiment evaluation, policy enforcement, and the categorization of news articles and academic research. For optimal performance, it is recommended to set the temperature to 0 and maintain a low output budget, which ensures that the results remain both consistent and easy to understand. Furthermore, Tev1 is recognized as a highly adaptable tool that meets a variety of classification requirements effectively and efficiently. Overall, its capabilities make it an essential asset in any decision-making framework.

Kimi K2.7 Code is an open-source agentic coding model from Moonshot AI designed for developers, engineering teams, and AI coding workflows that require long-context understanding and multi-step execution. It is built for real-world software engineering tasks, including code generation, code review, debugging, repository navigation, tool use, and long-horizon development work. The model is described by Moonshot AI as a coding-focused agentic model with stronger performance on complex coding tasks than earlier Kimi K2 releases. Kimi K2.7 Code supports a 256K context window, allowing it to process large codebases, technical requirements, logs, documentation, and multi-file development context in a single workflow. It is available through Kimi Code, which provides developer-oriented tools for using the model in coding tasks. The model can also be accessed through Moonshot’s API platform, where Kimi K2.7 Code and Kimi K2.7 Code Highspeed are offered alongside earlier Kimi models. For developers who want more control, Kimi K2.7 Code is listed on Hugging Face with deployment support for inference engines such as vLLM, SGLang, and KTransformers. It uses OpenAI- and Anthropic-compatible API options, helping teams connect it to existing applications, coding tools, and agent systems more easily. Third-party model listings describe it as using a 1T-parameter mixture-of-experts architecture with 32B active parameters, native INT4 quantization, and reduced thinking-token usage compared with Kimi K2.6. The model is designed to improve efficiency by using fewer reasoning tokens while still supporting demanding programming workflows. Kimi K2.7 Code is a strong fit for developers who want an open, long-context, tool-friendly AI model for software engineering automation and AI-assisted development.

Clef-flash is an advanced 9B multimodal decision-making framework designed to transform a given state and a structured set of inquiries into coherent decisions. It proficiently analyzes state data presented in various forms, including text, JSON, images, and video, offering a probability for each potential option related to the questions asked. With a context window accommodating up to 24,576 tokens, it can effortlessly process visual inputs, allowing developers to examine structured application data alongside visual elements in a unified manner. The model supports three distinct types of inquiries: "noul," which estimates the probability of a proposition's correctness; "choice," which allows users to select from a predefined list of alternatives, each with assigned probabilities; and "score," which evaluates inputs against a ranked set of criteria to produce a weighted probability score. Developers can submit between 1 and 64 questions in a single API request, with each response directly tied to its corresponding question ID. Its versatility makes it applicable for various scenarios, such as determining the urgency of support requests, routing them to the appropriate teams, and assessing customer impact severity, ultimately fostering enhanced operational efficiency in decision-making. By leveraging this model, organizations can refine their response strategies, optimize workflows, and significantly improve overall productivity in their operational processes.

Clef is an advanced multimodal decision-making model featuring 27 billion parameters, designed to convert a specified state and a schema of typed inquiries into organized results. This innovative model can process state information in multiple formats, such as text, JSON, images, and video, generating probabilities for each viable option associated with each inquiry. With a substantial context window of 65,536 tokens and the ability to process visual inputs, Clef allows developers to integrate both structured application data and visual content into a cohesive decision-making framework. It supports three unique formats for typed inquiries: noul, which evaluates the probability of a statement being accurate; choice, which facilitates selection from a set of predefined choices while offering probability estimates for each; and score, which measures an input against defined criteria to produce a probability-weighted score. Users can pose between 1 and 64 questions in a single request, with the answers being returned using the same identifiers, significantly improving the model's effectiveness and adaptability. This all-encompassing strategy guarantees that Clef can be utilized across various applications, catering to a wide spectrum of decision-making challenges and opportunities. Furthermore, the versatility of Clef's design makes it an invaluable tool for developers aiming to enhance their analytical capabilities in diverse fields.

Jev is a low-latency System One Model from TypeSafe AI built for structured decision-making rather than open-ended text generation. TypeSafe describes System One Models as a new class of frontier models intended to make fast decisions that can be consumed directly by application code. Jev takes unstructured information as input and returns predefined typed outputs with probabilities and confidence scores instead of generating unrestricted natural-language strings. Its parallel sampling architecture produces outputs in a single query rather than sequentially generating one token at a time, which is designed to reduce latency and inference cost. The model is trained with Reinforcement Learning for Calibrated Decisions, which focuses on producing well-calibrated probabilities for structured System One tasks. TypeSafe positions Jev as a software-native intelligence layer for tasks including classification, routing, scoring, extraction, decision branching, and other workflows that benefit from probabilistic logic. It can also be applied to model evaluation and safety workflows by scoring, judging, verifying, guardrailing, or detecting jailbreaks in prompts, reasoning traces, and generated outputs. The company reports service response times of roughly 70 to 500 milliseconds and says this can make Jev suitable for interactive and real-time applications where traditional frontier models may introduce too much latency. Jev is designed to return consistent schemas without type errors, allowing its results to be incorporated into code without the same parsing and validation steps commonly required for free-form LLM responses. TypeSafe also highlights map-reduce style processing over large datasets as a potential use case, where the model can turn large amounts of unstructured information into structured features and insights.

The Strands Decider is an efficient and agile decision-making framework, categorized as a System One model, specifically designed for applications in agency-driven AI. Unlike traditional language models that generate arbitrary text, this approach focuses on selecting from a predetermined set of options and efficiently answering yes-or-no questions or evaluating inputs on a ranked scale. Each decision rendered by the Strands Decider comes with a well-calibrated confidence score, which not only facilitates the swift handling of standard decisions but also allows for uncertain situations to be referred to a more advanced model or a human for in-depth examination. The first version of this model, named strands-decider-2B, consists of 1.9 billion parameters and is based on the Qwen3.5-2B-Base architecture, incorporating a refined pointer head instead of traditional language generation features. It executes evaluations in a single forward pass, thus bypassing the need for text generation or complex decoding processes. This model supports various types of decision-making, including option selections, noul for probability evaluations, and score for assessments on an ordered scale. Engineered for a wide range of purposes, it effectively aids in model routing, tool selection, validation of tool arguments, request triaging, establishing safety and grounding protocols, policy classification, and offering cost-effective evaluations, significantly broadening its applicability across numerous AI-driven tasks. Consequently, the Strands Decider stands out as a versatile tool that enhances operational efficiency in diverse AI workflows.

Laya AI embodies a collection of open-source, non-autoregressive decision-making models that prioritize swift and structured decision processes over the generation of free-form text. By analyzing a specific state alongside structured queries in one comprehensive step, it generates outputs that can be seamlessly integrated into application code. The system categorizes decisions into three types: choice, which selects from a set of predefined options and provides associated probabilities; score, which evaluates an input on a ranked scale such as urgency or severity; and noul, which predicts the likelihood that a statement is true. Designed for a wide range of tasks, Laya effectively manages agent and model routing, assists in ticket triage, and conducts risk assessments, among other functionalities. Its features enable the system to allocate requests to the correct department, gauge urgency levels, detect signs of potential refunds or customer churn, and identify instances of policy breaches or abuse. Additionally, it can escalate unclear cases to more sophisticated models or human experts for further examination. This adaptability makes Laya AI not only a valuable tool in operational settings but also enhances its ability to evolve with the changing demands of various industries. Ultimately, Laya AI distinguishes itself through its remarkable versatility and effectiveness in enhancing decision-making processes across a plethora of operational requirements.

d1 is the first decision model introduced by Liquid AI, which offers answers to structured questions about a given situation by utilizing probabilities rather than generating text responses. This innovative model efficiently processes unstructured data, such as textual and visual inputs, along with one or more inquiries in a single, cohesive forward pass, yielding probabilities without producing traditional output tokens. It supports three specific question formats: noul for simple yes-or-no choices, choice for selecting from a set of labels with their corresponding probabilities, and score for evaluating inputs on a ranked scale using probability metrics. Additionally, d1 is capable of processing multiple questions at once within a single query, thereby boosting its effectiveness in real-time decision-making scenarios. Liquid AI demonstrates the versatility of d1 across various applications, including the filtering of support tickets, exploring code repositories, intelligently categorizing documents into folders, managing websites, summarizing agent context, inspecting industrial parts, and even participating in games based on screenshots. This wide array of applications showcases not only the model's adaptability but also its potential to transform decision-making processes across different industries, emphasizing a notable leap forward in the realm of decision-making technologies.

The Decisions API is engineered to convert text and images into structured decisions that can be utilized by applications, boasting a processing speed approximately ten times faster than the Responses API. To use it effectively, developers are required to submit a model along with shared input evidence and one or more questions that define the evaluation parameters, after which the API provides organized responses that are ideal for various tasks such as content classification, request routing, prioritizing work, or making decisions based on established thresholds. It supports three types of inquiries: predicate questions, which determine the truth of a condition and yield a probability between 0 and 1; choice questions, which select a single option from a specified set while offering corresponding probabilities and confidence levels; and score questions, which evaluate an input against a series of ordered rubric levels, resulting in a probability-weighted score. The API's flexibility allows it to handle inputs as simple text or as user messages that incorporate both text and images, making it suitable for a wide array of scenarios including detecting visible product damage, interpreting request intent, classifying various document types, assessing the severity of issues, or gauging answer relevance. By enabling this level of processing, the system significantly enhances the ability of developers to build applications that are not only more responsive but also smarter, ultimately catering better to the needs of users. Furthermore, this advanced capability means that applications can adapt to changing user requirements in real-time, further increasing their effectiveness.

Microsoft-Decision-1 is a highly effective decision-scoring model specifically designed for structured outputs, allowing software systems to take swift actions. This innovative model supports a range of functionalities including routing, classification, prioritization, verification, and workflow management, applicable across various applications, agents, and automated systems. Developed through post-training of Qwen3.5-9B, it excels in single-pass decision scoring and delivers calibrated probabilities for a defined set of response choices. The model is adept at addressing yes/no inquiries, multiple-choice situations, rating scales, and rubric-based assessments of AI-generated outputs and agent performances via a structured API. Comprehensive evaluations carried out by Microsoft across 36 benchmarks covering routing, ranking, long-context understanding, multilingual functions, and tasks outside conventional distributions demonstrate that it achieved top accuracy levels with impressively low latency. Additionally, the model places a strong emphasis on robustness, with extensive testing conducted to ensure decision-making stability even when prompts undergo rephrasing or modifications. This commitment to stability highlights its potential for delivering reliable performance in diverse real-world scenarios, making it a versatile tool for various industries.

Perplexity's Decisions API is crafted for swift and organized decision-making, making it readily applicable for software solutions. Instead of generating unstructured text, it processes diverse inputs like natural-language sentences, JSON data, or images, creating a unified context from which it offers structured, typed responses that come with associated probabilities. Developers can enhance their queries by adding multiple named questions to a single input, with the option to choose from three unique question formats. The "noul" type facilitates straightforward yes-or-no assessments, providing a probability score for a "yes" answer that ranges from 0 to 1. The "choice" type enables the specification of different options, returning the most likely choice along with the associated probabilities for each alternative and a measure of confidence. The "score" type, on the other hand, evaluates content against a set rubric, yielding a probability-weighted score, a comprehensive probability distribution, and an associated confidence level. This API is well-suited for numerous applications, such as classification, routing, rubric-based grading, severity assessments, sentiment analysis, and any decision-making process that can be measured against a defined threshold, thus positioning it as an essential tool for developers. Furthermore, its systematic methodology not only increases the dependability of the decisions generated by the software but also guarantees that the results are both precise and stable, enhancing the overall user experience. Such reliability in decision-making can significantly contribute to the success of various projects.

Perplexity Decider represents an advanced decision-making model featuring 27 billion parameters, evolved from Qwen3.8-27B, and is designed specifically for structured classification and multimodal decision-making tasks. Rather than generating freeform text, it evaluates an input state alongside a predefined decision schema to present the selected outcome along with calibrated probabilities. This model excels in making choice decisions from established categories and also handles noul decisions, assessing the probability of a yes-or-no answer. Additionally, it can analyze images in combination with text, allowing it to make well-informed decisions based on visual inputs such as screenshots. Common use cases involve routing support requests to the correct departments, like billing, technical support, or sales, and determining whether a message conveys urgency. With an overall accuracy rate of 85.71% across 11 diverse benchmarks, including FinancialPhraseBank, RAGTruth, JudgeBench, TabFact, ContractNLI, Circa, Belebele, and TruthfulQA binary, Perplexity Decider illustrates remarkable versatility and effectiveness in various decision-making environments. This significant accuracy level not only highlights the model's capabilities but also points to its potential for improving automated systems that require a nuanced understanding of both text and images, paving the way for future advancements in decision-making technologies.

Span-01 represents a cutting-edge hyper-parallel reasoning classifier designed to detect complex patterns within large production traces in a single forward pass. Rather than generating tokens sequentially, it evaluates natural-language behavior definitions and outputs direct probabilities that indicate whether a specific behavior is present, absent, or unobservable. The model begins with a foundational training phase focused on general classification reasoning through Reinforcement Learning with Human Feedback (RLAIF), which is followed by fine-tuning aimed explicitly at behavior detection. This training methodology empowers Span-01 to effectively leverage previously unseen definitions for both new behaviors and diverse domains, thereby circumventing the constraints of a fixed taxonomy. During inference, its hyper-parallel definition branches are capable of assessing multiple behaviors at once against the same trace context. Additionally, the use of hybrid attention enhances its ability to reason over lengthy and complex inputs. Engineered for the scrutiny of production environments, Span-01 emphasizes safety, reliability, grounding, agent behavior, and response quality while tackling challenges like jailbreak and prompt injection detection alongside privacy and confidentiality issues. This innovative methodology not only makes Span-01 a noteworthy milestone in the domain but also paves the way for more efficient and thorough investigations into intricate behaviors, ultimately leading to improved decision-making processes. As a result, its deployment could significantly enhance operational standards across various industries.

Solar Decide serves as Upstage's decision-making framework, functioning as a System One endpoint on Solar Mini 4. Instead of generating text, it evaluates a specific state alongside one or more designated questions to produce a selection, score, or a binary yes/no answer, complete with calibrated probabilities that stem from the model itself. Each question is processed in a single forward pass, which significantly boosts the model's efficiency for various tasks, including quick classification, routing, policy validation, prioritization, and other structured decision-making processes. Users can opt for a choice type that allows them to pick from a list of predefined alternatives, yielding not only the selected option but also the related probabilities and levels of confidence. On the other hand, the score type assesses an input based on particular ranked criteria, delivering a probability-weighted score along with an explanatory legend, probability distribution, and metrics of confidence. Moreover, the noul type evaluates the probability that a yes-or-no statement is correct. Users can also pose multiple questions regarding the same state in a single request, where the state can be presented as plain text or formatted as any JSON value for the prompt. This adaptability makes Solar Decide an invaluable resource for effectively navigating intricate decision-making scenarios, ultimately enhancing the decision-making process across various contexts.

Big Pickle is an AI model available through OpenCode Zen, a provider that curates and validates models for coding-agent use cases. The model is listed under the OpenCode provider and can be accessed through an OpenAI-compatible completions API. Big Pickle supports text input and reasoning, making it suitable for developer workflows that require analysis, planning, code understanding, and multi-step execution. It is also described as supporting function calling, which helps developers connect model output with tools, agents, scripts, and automated workflows. Big Pickle’s large context window makes it useful for working with extended prompts, larger project files, documentation, codebases, and complex technical tasks. The model appears in OpenCode Zen’s model list alongside other coding and reasoning models, positioning it as part of a developer-focused model ecosystem. Third-party model directories list Big Pickle with free input and output token pricing, making it appealing for experimentation and cost-sensitive workloads. Developers can use Big Pickle for code assistance, refactoring, debugging, technical research, task decomposition, command-line workflows, and AI agent orchestration. Because some listings differ on exact output-token limits, teams should verify the current model configuration directly in their OpenCode environment before designing production workloads around a fixed limit. Big Pickle is especially useful for developers who want to test long-context AI coding workflows without committing to a more expensive model tier. Big Pickle helps engineering teams explore AI-assisted development, coding agents, tool calling, and long-context reasoning in a flexible and accessible way.

LongCat-2.0 signifies a remarkable leap forward in the field of language models, boasting an impressive 1.6 trillion parameters through a Mixture-of-Experts architecture that utilizes AI ASIC superpods, with around 48 billion parameters activated per token, demonstrating outstanding proficiency in coding and agentic functions. This model notably surpasses its predecessors by incorporating a large-scale sparse architecture along with specialized post-training techniques designed specifically for applications in real-world software development, tool usage, long-context reasoning, and intricate agent operations. Entirely built and executed on AI ASIC superpods, LongCat-2.0's pretraining involved processing over 35 trillion tokens and countless accelerator hours, highlighting the forefront of training techniques on state-of-the-art hardware. To further enhance its capabilities on tasks that require long-term contextual awareness, the model integrates LongCat Sparse Attention and is trained with hundreds of billions of tokens derived from 1M-context datasets, which empowers it to adeptly handle ultra-long context challenges and maintain a comprehensive understanding of extensive documents. This unique blend of features not only establishes LongCat-2.0 as an innovative leader in advanced language models but also sets a new benchmark for future developments in the domain. Its capabilities are likely to inspire a new wave of research and applications in the field.

SubQ 1.1 Small is a long-context enterprise AI model developed by Subquadratic to address the limitations of traditional models that struggle with large artifacts. It is built for tasks where the full context matters, including analyzing entire codebases, reviewing lengthy contracts, comparing financial filings, and reasoning across document collections. The model uses Subquadratic Sparse Attention, which replaces dense attention with a learned sparse approach that scales more efficiently as context length grows. This allows SubQ 1.1 Small to process extremely large context windows while sharply reducing attention compute requirements. In benchmark testing, the model achieved near-perfect needle-in-a-haystack retrieval at 1M, 2M, 6M, and 12M tokens. It also scored 99.12% on the RULER 128K benchmark, demonstrating strength on tasks involving multi-hop reasoning, variable tracing, aggregation, and long-context understanding. Beyond retrieval, SubQ 1.1 Small maintains competitive performance in general knowledge, coding, and enterprise agent benchmarks such as GPQA Diamond, LiveCodeBench, and AutomationBench Finance. Its efficiency is a major advantage, requiring 64.5x less compute than dense attention and running 56x faster than FlashAttention-2 at 1M tokens on a single attention layer. The model was trained through staged context extension and continued pretraining on long-form artifacts such as books, documents, and repository-scale code. SubQ 1.1 Small is suited for financial analysis, legal work, software engineering, due diligence, long-horizon coding tasks, and enterprise workflows that depend on relationships spread across large bodies of information. It gives organizations a way to reason over complete artifacts more directly instead of relying only on retrieval pipelines, chunking strategies, and agentic scaffolding.

SubQ is a next-generation large language model developed by Subquadratic, designed to handle extremely long-context reasoning tasks with high efficiency. It supports up to 12 million tokens in a single prompt, allowing it to process entire codebases, months of development history, and large datasets in one step. The model uses a fully sub-quadratic sparse-attention architecture, which reduces unnecessary computations by focusing only on meaningful relationships between data points. This approach significantly lowers computational costs while maintaining strong performance across complex tasks. SubQ is optimized for use cases such as software engineering, code analysis, long-context retrieval, and AI agent workflows. It enables developers to analyze large amounts of information without breaking it into smaller segments. The model offers fast processing speeds and lower operational costs compared to traditional transformer-based models. SubQ is accessible through APIs, making it easy for developers and enterprises to integrate it into their systems. It can also be used within coding agents to improve code mapping, exploration, and understanding. The platform supports streaming and tool usage for more dynamic workflows. Its architecture allows it to scale efficiently as data size increases, overcoming common limitations of standard models. SubQ also delivers competitive performance on benchmarks related to coding and long-context tasks. By combining efficiency, scalability, and large context capabilities, it provides a powerful solution for advanced AI applications.

Atlas is a cutting-edge omni world model crafted for spatial intelligence, effortlessly interacting with text, images, video, and 3D data. As a sophisticated multimodal autoregressive diffusion transformer, it harmonizes diverse inputs into a unified spatial framework, while simultaneously forecasting subsequent elements and maintaining coherence in three-dimensional spaces through its insights and creative ideas. This robust model enables world creation, reconstruction, and simulation across a wide array of applications. It is adept at generating images and videos from one or several reference images, providing meticulous camera control to produce extensive and coherent videos that adhere to custom-designed camera paths. In the realm of spatial reconstruction, Atlas shines in reinterpreting real-world settings from a limited number of input images, which allows it to create distinct viewpoints and deliver detailed 3D outputs such as point clouds and 3D Gaussian splats. With the incorporation of additional input perspectives, the model can further improve contextual understanding, resulting in progressively precise reconstructions with a diminished dependence on imagination. Additionally, Atlas has the capability to analyze and forecast the progression of worlds over time, thereby adding a dynamic aspect to its functionality, which enhances its versatility in various scenarios. This multifaceted approach not only enriches the user experience but also expands the potential applications of the model across different fields.

GLM-5 is Z.ai’s most advanced open-source model to date, purpose-built for complex systems engineering, long-horizon planning, and autonomous agent workflows. Building on the foundation of GLM-4.5, it dramatically scales both total parameters and pre-training data while increasing active parameter efficiency. The integration of DeepSeek Sparse Attention allows GLM-5 to maintain strong long-context reasoning capabilities while reducing deployment costs. To improve post-training performance, Z.ai developed slime, an asynchronous reinforcement learning infrastructure that significantly boosts training throughput and iteration speed. As a result, GLM-5 achieves top-tier performance among open-source models across reasoning, coding, and general agent benchmarks. It demonstrates exceptional strength in long-term operational simulations, including leading results on Vending Bench 2, where it manages a year-long simulated business with strong financial outcomes. In coding evaluations such as SWE-bench and Terminal-Bench 2.0, GLM-5 delivers competitive results that narrow the gap with proprietary frontier systems. The model is fully open-sourced under the MIT License and available through Hugging Face, ModelScope, and Z.ai’s developer platforms. Developers can deploy GLM-5 locally using inference frameworks like vLLM and SGLang, including support for non-NVIDIA hardware through optimization and quantization techniques. Through Z.ai, users can access both Chat Mode for fast interactions and Agent Mode for tool-augmented, multi-step task execution. GLM-5 also enables structured document generation, producing ready-to-use .docx, .pdf, and .xlsx files for business and academic workflows. With compatibility across coding agents and cross-application automation frameworks, GLM-5 moves foundation models from conversational assistants toward full-scale work engines.

We are excited to present DeepSeek-V3.2-Exp, our latest experimental model that evolves from V3.1-Terminus, incorporating the cutting-edge DeepSeek Sparse Attention (DSA) technology designed to significantly improve both training and inference speeds for longer contexts. This innovative DSA framework enables accurate sparse attention while preserving the quality of outputs, resulting in enhanced performance for long-context tasks alongside reduced computational costs. Benchmark evaluations demonstrate that V3.2-Exp delivers performance on par with V3.1-Terminus, all while benefiting from these efficiency gains. The model is fully functional across various platforms, including app, web, and API. In addition, to promote wider accessibility, we have reduced DeepSeek API pricing by more than 50% starting now. During this transition phase, users will have access to V3.1-Terminus through a temporary API endpoint until October 15, 2025. DeepSeek invites feedback on DSA from users via our dedicated feedback portal, encouraging community engagement. To further support this initiative, DeepSeek-V3.2-Exp is now available as open-source, with model weights and key technologies—including essential GPU kernels in TileLang and CUDA—published on Hugging Face, and we are eager to observe how the community will leverage this significant technological advancement. As we unveil this new chapter, we anticipate fruitful interactions and innovative applications arising from the collective contributions of our user base.

Composer is an AI model developed by Cursor, specifically designed for software engineering tasks, providing fast and interactive coding assistance within the Cursor IDE, an upgraded version of a VS Code-based editor that features intelligent automation capabilities. This model uses a mixture-of-experts framework and reinforcement learning (RL) to address real-world coding challenges encountered in large codebases, allowing it to offer quick, contextually relevant responses that include code adjustments, planning, and insights into project frameworks, tools, and conventions, achieving generation speeds that are nearly four times faster than those of its peers in performance evaluations. With a focus on the development workflow, Composer incorporates long-context understanding, semantic search functionalities, and limited tool access (including file manipulation and terminal commands) to effectively resolve complex engineering questions with practical and efficient solutions. Its distinctive architecture not only enables adaptability across various programming environments but also ensures that users receive personalized support tailored to their individual coding requirements. Furthermore, the versatility of Composer allows it to evolve alongside the ever-changing landscape of software development, making it an invaluable resource for developers seeking to enhance their coding experience.

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.

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.

DeepSeek-V4 is a cutting-edge open-source AI model built to deliver exceptional performance in reasoning, coding, and large-scale data processing. It supports an industry-leading one million token context window, allowing it to manage long documents and complex tasks efficiently. The model includes two variants: DeepSeek-V4-Pro, which offers 1.6 trillion parameters with 49 billion active for top-tier performance, and DeepSeek-V4-Flash, which provides a faster and more cost-effective alternative. DeepSeek-V4 introduces structural innovations such as token-wise compression and sparse attention, significantly reducing computational overhead while maintaining accuracy. It is designed with strong agentic capabilities, enabling seamless integration with AI agents and multi-step workflows. The model excels in domains such as mathematics, coding, and scientific reasoning, outperforming many open-source alternatives. It also supports flexible reasoning modes, allowing users to optimize for speed or depth depending on the task. DeepSeek-V4 is compatible with popular APIs, making it easy to integrate into existing systems. Its open-source nature allows developers to customize and scale it according to their needs. The model is already being used in advanced coding agents and automation workflows. It delivers a strong balance of performance, efficiency, and scalability for real-world applications. Overall, DeepSeek-V4 represents a major advancement in accessible, high-performance AI technology.

Oxlo.ai presents a privacy-focused inference platform specifically designed for agents, enabling the use of advanced open-source models while guaranteeing unrestricted agentic tool access, reliable failover options, and no data retention or training. Developers can take advantage of request-based access to a variety of carefully selected open models through a simplified HTTP API, ensuring predictable usage, low-latency inference, and smooth integration with existing production systems. Teams can conveniently call models using endpoints compatible with OpenAI, switch from other service providers with just a modification of the base URL and API key, and enjoy ongoing support for several features such as streaming, function calling, JSON mode, and a variety of model types that include vision models, embeddings, and image generation capabilities. With compatibility for over 40 distinct models, Oxlo.ai supports a comprehensive range of applications, including text, chat, reasoning, coding, image generation, audio processing, embeddings, computer vision, vision-language tasks, speech-to-text, text-to-speech, long-context handling, and detection workflows, establishing it as a flexible resource for developers. This broad support fosters innovative applications across various sectors, significantly improving the potential of teams eager to utilize state-of-the-art AI technologies and pushing the boundaries of what's possible in their projects. By integrating Oxlo.ai into their workflows, organizations can harness the power of advanced AI while maintaining a strong commitment to user privacy.