Alternatives to Olmo 3

The MiniMax‑M1 model, introduced by MiniMax AI and licensed under Apache 2.0, represents a significant advancement in hybrid-attention reasoning architecture. With an extraordinary capacity for handling a 1 million-token context window and generating outputs of up to 80,000 tokens, it facilitates in-depth analysis of lengthy texts. Utilizing a cutting-edge CISPO algorithm, MiniMax‑M1 was trained through extensive reinforcement learning, achieving completion on 512 H800 GPUs in approximately three weeks. This model sets a new benchmark in performance across various domains, including mathematics, programming, software development, tool utilization, and understanding of long contexts, either matching or surpassing the capabilities of leading models in the field. Additionally, users can choose between two distinct variants of the model, each with a thinking budget of either 40K or 80K, and access the model's weights and deployment instructions on platforms like GitHub and Hugging Face. Such features make MiniMax‑M1 a versatile tool for developers and researchers alike.

Qwen3-Max represents Alibaba's cutting-edge large language model, featuring a staggering trillion parameters aimed at enhancing capabilities in tasks that require agency, coding, reasoning, and managing lengthy contexts. This model is an evolution of the Qwen3 series, leveraging advancements in architecture, training methods, and inference techniques; it integrates both thinker and non-thinker modes, incorporates a unique “thinking budget” system, and allows for dynamic mode adjustments based on task complexity. Capable of handling exceptionally lengthy inputs, processing hundreds of thousands of tokens, it also supports tool invocation and demonstrates impressive results across various benchmarks, including coding, multi-step reasoning, and agent evaluations like Tau2-Bench. While the initial version prioritizes instruction adherence in a non-thinking mode, Alibaba is set to introduce reasoning functionalities that will facilitate autonomous agent operations in the future. In addition to its existing multilingual capabilities and extensive training on trillions of tokens, Qwen3-Max is accessible through API interfaces that align seamlessly with OpenAI-style functionalities, ensuring broad usability across applications. This comprehensive framework positions Qwen3-Max as a formidable player in the realm of advanced artificial intelligence language models.

GLM-5 is a next-generation open-source foundation model from Z.ai designed to push the boundaries of agentic engineering and complex task execution. Compared to earlier versions, it significantly expands parameter count and training data, while introducing DeepSeek Sparse Attention to optimize inference efficiency. The model leverages a novel asynchronous reinforcement learning framework called slime, which enhances training throughput and enables more effective post-training alignment. GLM-5 delivers leading performance among open-source models in reasoning, coding, and general agent benchmarks, with strong results on SWE-bench, BrowseComp, and Vending Bench 2. Its ability to manage long-horizon simulations highlights advanced planning, resource allocation, and operational decision-making skills. Beyond benchmark performance, GLM-5 supports real-world productivity by generating fully formatted documents such as .docx, .pdf, and .xlsx files. It integrates with coding agents like Claude Code and OpenClaw, enabling cross-application automation and collaborative agent workflows. Developers can access GLM-5 via Z.ai’s API, deploy it locally with frameworks like vLLM or SGLang, or use it through an interactive GUI environment. The model is released under the MIT License, encouraging broad experimentation and adoption. Overall, GLM-5 represents a major step toward practical, work-oriented AI systems that move beyond chat into full task execution.

OLMo 2 represents a collection of completely open language models created by the Allen Institute for AI (AI2), aimed at giving researchers and developers clear access to training datasets, open-source code, reproducible training methodologies, and thorough assessments. These models are trained on an impressive volume of up to 5 trillion tokens and compete effectively with top open-weight models like Llama 3.1, particularly in English academic evaluations. A key focus of OLMo 2 is on ensuring training stability, employing strategies to mitigate loss spikes during extended training periods, and applying staged training interventions in the later stages of pretraining to mitigate weaknesses in capabilities. Additionally, the models leverage cutting-edge post-training techniques derived from AI2's Tülu 3, leading to the development of OLMo 2-Instruct models. To facilitate ongoing enhancements throughout the development process, an actionable evaluation framework known as the Open Language Modeling Evaluation System (OLMES) was created, which includes 20 benchmarks that evaluate essential capabilities. This comprehensive approach not only fosters transparency but also encourages continuous improvement in language model performance.

DeepSeek-V3.2 is a highly optimized large language model engineered to balance top-tier reasoning performance with significant computational efficiency. It builds on DeepSeek's innovations by introducing DeepSeek Sparse Attention (DSA), a custom attention algorithm that reduces complexity and excels in long-context environments. The model is trained using a sophisticated reinforcement learning approach that scales post-training compute, enabling it to perform on par with GPT-5 and match the reasoning skill of Gemini-3.0-Pro. Its Speciale variant overachieves in demanding reasoning benchmarks and does not include tool-calling capabilities, making it ideal for deep problem-solving tasks. DeepSeek-V3.2 is also trained using an agentic synthesis pipeline that creates high-quality, multi-step interactive data to improve decision-making, compliance, and tool-integration skills. It introduces a new chat template design featuring explicit thinking sections, improved tool-calling syntax, and a dedicated developer role used strictly for search-agent workflows. Users can encode messages using provided Python utilities that convert OpenAI-style chat messages into the expected DeepSeek format. Fully open-source under the MIT license, DeepSeek-V3.2 is a flexible, cutting-edge model for researchers, developers, and enterprise AI teams.

DeepSeek-V3.2-Speciale is the most advanced reasoning-focused version of the DeepSeek-V3.2 family, designed to excel in mathematical, algorithmic, and logic-intensive tasks. It incorporates DeepSeek Sparse Attention (DSA), an efficient attention mechanism tailored for very long contexts, enabling scalable reasoning with minimal compute costs. The model undergoes a robust reinforcement learning pipeline that scales post-training compute to frontier levels, enabling performance that exceeds GPT-5 on internal evaluations. Its achievements include gold-medal-level solutions in IMO 2025, IOI 2025, ICPC World Finals, and CMO 2025, with final submissions publicly released for verification. Unlike the standard V3.2 model, the Speciale variant removes tool-calling capabilities to maximize focused reasoning output without external interactions. DeepSeek-V3.2-Speciale uses a revised chat template with explicit thinking blocks and system-level reasoning formatting. The repository includes encoding tools showing how to convert OpenAI-style chat messages into DeepSeek’s specialized input format. With its MIT license and 685B-parameter architecture, DeepSeek-V3.2-Speciale offers cutting-edge performance for academic research, competitive programming, and enterprise-level reasoning applications.

Kimi K2 represents a cutting-edge series of open-source large language models utilizing a mixture-of-experts (MoE) architecture, with a staggering 1 trillion parameters in total and 32 billion activated parameters tailored for optimized task execution. Utilizing the Muon optimizer, it has been trained on a substantial dataset of over 15.5 trillion tokens, with its performance enhanced by MuonClip’s attention-logit clamping mechanism, resulting in remarkable capabilities in areas such as advanced knowledge comprehension, logical reasoning, mathematics, programming, and various agentic operations. Moonshot AI offers two distinct versions: Kimi-K2-Base, designed for research-level fine-tuning, and Kimi-K2-Instruct, which is pre-trained for immediate applications in chat and tool interactions, facilitating both customized development and seamless integration of agentic features. Comparative benchmarks indicate that Kimi K2 surpasses other leading open-source models and competes effectively with top proprietary systems, particularly excelling in coding and intricate task analysis. Furthermore, it boasts a generous context length of 128 K tokens, compatibility with tool-calling APIs, and support for industry-standard inference engines, making it a versatile option for various applications. The innovative design and features of Kimi K2 position it as a significant advancement in the field of artificial intelligence language processing.

CodeGemma represents an impressive suite of efficient and versatile models capable of tackling numerous coding challenges, including middle code completion, code generation, natural language processing, mathematical reasoning, and following instructions. It features three distinct model types: a 7B pre-trained version designed for code completion and generation based on existing code snippets, a 7B variant fine-tuned for translating natural language queries into code and adhering to instructions, and an advanced 2B pre-trained model that offers code completion speeds up to twice as fast. Whether you're completing lines, developing functions, or crafting entire segments of code, CodeGemma supports your efforts, whether you're working in a local environment or leveraging Google Cloud capabilities. With training on an extensive dataset comprising 500 billion tokens predominantly in English, sourced from web content, mathematics, and programming languages, CodeGemma not only enhances the syntactical accuracy of generated code but also ensures its semantic relevance, thereby minimizing mistakes and streamlining the debugging process. This powerful tool continues to evolve, making coding more accessible and efficient for developers everywhere.

DeepSeek-V4 is an advanced open large language model engineered for high-efficiency reasoning, sophisticated problem solving, and powerful agent-based execution. At its core is DeepSeek Sparse Attention (DSA), a specialized long-context attention mechanism that minimizes computational costs without sacrificing accuracy or depth. The model leverages a scalable reinforcement learning framework to refine reasoning quality and align outputs with real-world task demands. A dedicated agent task synthesis pipeline generates structured reasoning traces and tool-use demonstrations, strengthening post-training performance. DeepSeek-V4 features an updated chat architecture with improved tool-calling logic designed for multi-step workflows. The introduction of an optional developer role enhances orchestration in agent-driven environments. Its architecture supports extended context handling for research-intensive and enterprise applications. Optimized for both experimentation and deployment, the model balances efficiency with frontier-level capability. DeepSeek-V4 stands out as a competitive open alternative for advanced AI reasoning and autonomous task execution.

Molmo 2 represents a cutting-edge suite of open vision-language models that come with completely accessible weights, training data, and code, thereby advancing the original Molmo series' capabilities in grounded image comprehension to encompass video and multiple image inputs. This evolution enables sophisticated video analysis, including pointing, tracking, dense captioning, and question-answering functionalities, all of which demonstrate robust spatial and temporal reasoning across frames. The suite consists of three distinct models: an 8 billion-parameter variant tailored for comprehensive video grounding and QA tasks, a 4 billion-parameter model that prioritizes efficiency, and a 7 billion-parameter model backed by Olmo, which features a fully open end-to-end architecture that includes the foundational language model. Notably, these new models surpass their predecessors on key benchmarks, setting unprecedented standards for open-model performance in image and video comprehension tasks. Furthermore, they often rival significantly larger proprietary systems while being trained on a much smaller dataset compared to similar closed models, showcasing their efficiency and effectiveness in the field. This impressive achievement marks a significant advancement in the accessibility and performance of AI-driven visual understanding technologies.

DeepSeek-Coder-V2 is an open-source model tailored for excellence in programming and mathematical reasoning tasks. Utilizing a Mixture-of-Experts (MoE) architecture, it boasts a staggering 236 billion total parameters, with 21 billion of those being activated per token, which allows for efficient processing and outstanding performance. Trained on a massive dataset comprising 6 trillion tokens, this model enhances its prowess in generating code and tackling mathematical challenges. With the ability to support over 300 programming languages, DeepSeek-Coder-V2 has consistently outperformed its competitors on various benchmarks. It is offered in several variants, including DeepSeek-Coder-V2-Instruct, which is optimized for instruction-based tasks, and DeepSeek-Coder-V2-Base, which is effective for general text generation. Additionally, the lightweight options, such as DeepSeek-Coder-V2-Lite-Base and DeepSeek-Coder-V2-Lite-Instruct, cater to environments that require less computational power. These variations ensure that developers can select the most suitable model for their specific needs, making DeepSeek-Coder-V2 a versatile tool in the programming landscape.

Xgen-small is a compact language model crafted by Salesforce AI Research that is tailored for enterprise use, offering efficient long-context capabilities at a manageable cost. It employs a combination of focused data curation, scalable pre-training, length extension, instruction fine-tuning, and reinforcement learning to address the intricate and high-volume inference needs of contemporary businesses. In contrast to conventional large models, Xgen-small excels in processing extensive contexts, allowing it to effectively synthesize insights from various sources such as internal documents, code bases, research articles, and real-time data feeds. With parameter sizes of 4B and 9B, it strikes a careful balance between cost efficiency, privacy protections, and comprehensive long-context comprehension, positioning itself as a reliable and sustainable option for large-scale Enterprise AI implementation. This innovative approach not only enhances operational efficiency but also empowers organizations to leverage AI effectively in their strategic initiatives.

StableCode provides an innovative solution for developers aiming to enhance their productivity through the utilization of three distinct models designed to assist in coding tasks. Initially, the foundational model was developed using a broad range of programming languages sourced from the stack-dataset (v1.2) by BigCode, with subsequent training focused on widely-used languages such as Python, Go, Java, JavaScript, C, Markdown, and C++. In total, our models have been trained on an impressive 560 billion tokens of code using our high-performance computing cluster. Once the base model was created, an instruction model was meticulously fine-tuned for particular use cases, enabling it to tackle intricate programming challenges effectively. To achieve this refinement, approximately 120,000 pairs of code instructions and responses in Alpaca format were utilized to train the base model. StableCode serves as a perfect foundation for those eager to deepen their understanding of programming, while the long-context window model provides an exceptional assistant that delivers both single-line and multi-line autocomplete suggestions seamlessly. This advanced model is specifically designed to efficiently manage larger chunks of code simultaneously, enhancing the overall coding experience for developers. By integrating these features, StableCode not only aids in coding but also fosters a deeper learning environment for aspiring programmers.

Seed2.0 Lite belongs to the Seed2.0 lineup from ByteDance, which encompasses versatile multimodal AI agent models engineered to tackle intricate, real-world challenges while maintaining a harmonious balance between efficiency and performance. This model boasts superior multimodal comprehension and instruction-following skills compared to its predecessors in the Seed series, allowing it to effectively interpret and analyze text, visual components, and structured data for use in production environments. Positioned as a mid-sized option within the family, Lite is fine-tuned to provide high-quality results with quick responsiveness at a reduced cost and faster inference times than the Pro version, while also enhancing the capabilities of earlier models. Consequently, it is well-suited for applications that demand consistent reasoning, extended context comprehension, and the execution of multimodal tasks without necessitating the utmost raw performance levels. Moreover, this accessibility makes Seed2.0 Lite an attractive choice for developers seeking efficiency alongside capabilities in their AI solutions.

Phi-4-mini-flash-reasoning is a 3.8 billion-parameter model that is part of Microsoft's Phi series, specifically designed for edge, mobile, and other environments with constrained resources where processing power, memory, and speed are limited. This innovative model features the SambaY hybrid decoder architecture, integrating Gated Memory Units (GMUs) with Mamba state-space and sliding-window attention layers, achieving up to ten times the throughput and a latency reduction of 2 to 3 times compared to its earlier versions without compromising on its ability to perform complex mathematical and logical reasoning. With a support for a context length of 64K tokens and being fine-tuned on high-quality synthetic datasets, it is particularly adept at handling long-context retrieval, reasoning tasks, and real-time inference, all manageable on a single GPU. Available through platforms such as Azure AI Foundry, NVIDIA API Catalog, and Hugging Face, Phi-4-mini-flash-reasoning empowers developers to create applications that are not only fast but also scalable and capable of intensive logical processing. This accessibility allows a broader range of developers to leverage its capabilities for innovative solutions.

Gemini 2.5 Pro Deep Think is the latest evolution of Google’s Gemini models, specifically designed to tackle more complex tasks with better accuracy and efficiency. The key feature of Deep Think enables the AI to think through its responses, improving its reasoning and enhancing decision-making processes. This model is a game-changer for coding, problem-solving, and AI-driven conversations, with support for multimodality, long context windows, and advanced coding capabilities. It integrates native audio outputs for richer, more expressive interactions and is optimized for speed and accuracy across various benchmarks. With the addition of this advanced reasoning mode, Gemini 2.5 Pro Deep Think is not just faster but also smarter, handling complex queries with ease.

Introducing the next iteration of our open-source large language model, this version features model weights along with initial code for the pretrained and fine-tuned Llama language models, which span from 7 billion to 70 billion parameters. The Llama 2 pretrained models have been developed using an impressive 2 trillion tokens and offer double the context length compared to their predecessor, Llama 1. Furthermore, the fine-tuned models have been enhanced through the analysis of over 1 million human annotations. Llama 2 demonstrates superior performance against various other open-source language models across multiple external benchmarks, excelling in areas such as reasoning, coding capabilities, proficiency, and knowledge assessments. For its training, Llama 2 utilized publicly accessible online data sources, while the fine-tuned variant, Llama-2-chat, incorporates publicly available instruction datasets along with the aforementioned extensive human annotations. Our initiative enjoys strong support from a diverse array of global stakeholders who are enthusiastic about our open approach to AI, including companies that have provided valuable early feedback and are eager to collaborate using Llama 2. The excitement surrounding Llama 2 signifies a pivotal shift in how AI can be developed and utilized collectively.

Tencent Hunyuan represents a comprehensive family of multimodal AI models crafted by Tencent, encompassing a range of modalities including text, images, video, and 3D data, all aimed at facilitating general-purpose AI applications such as content creation, visual reasoning, and automating business processes. This model family features various iterations tailored for tasks like natural language interpretation, multimodal comprehension that combines vision and language (such as understanding images and videos), generating images from text, creating videos, and producing 3D content. The Hunyuan models utilize a mixture-of-experts framework alongside innovative strategies, including hybrid "mamba-transformer" architectures, to excel in tasks requiring reasoning, long-context comprehension, cross-modal interactions, and efficient inference capabilities. A notable example is the Hunyuan-Vision-1.5 vision-language model, which facilitates "thinking-on-image," allowing for intricate multimodal understanding and reasoning across images, video segments, diagrams, or spatial information. This robust architecture positions Hunyuan as a versatile tool in the rapidly evolving field of AI, capable of addressing a diverse array of challenges.

The Pro version of OpenAI’s latest GPT-5.2 model family, known as GPT-5.2 Pro, stands out as the most advanced offering, designed to provide exceptional reasoning capabilities, tackle intricate tasks, and achieve heightened accuracy suitable for high-level knowledge work, innovative problem-solving, and enterprise applications. Building upon the enhancements of the standard GPT-5.2, it features improved general intelligence, enhanced understanding of longer contexts, more reliable factual grounding, and refined tool usage, leveraging greater computational power and deeper processing to deliver thoughtful, dependable, and contextually rich responses tailored for users with complex, multi-step needs. GPT-5.2 Pro excels in managing demanding workflows, including sophisticated coding and debugging, comprehensive data analysis, synthesis of research, thorough document interpretation, and intricate project planning, all while ensuring greater accuracy and reduced error rates compared to its less robust counterparts. This makes it an invaluable tool for professionals seeking to optimize their productivity and tackle substantial challenges with confidence.

GigaChat 3 Ultra redefines open-source scale by delivering a 702B-parameter frontier model purpose-built for Russian and multilingual understanding. Designed with a modern MoE architecture, it achieves the reasoning strength of giant dense models while using only a fraction of active parameters per generation step. Its massive 14T-token training corpus includes natural human text, curated multilingual sources, extensive STEM materials, and billions of high-quality synthetic examples crafted to boost logic, math, and programming skills. This model is not a derivative or retrained foreign LLM—it is a ground-up build engineered to capture cultural nuance, linguistic accuracy, and reliable long-context performance. GigaChat 3 Ultra integrates seamlessly with open-source tooling like vLLM, sglang, DeepSeek-class architectures, and HuggingFace-based training stacks. It supports advanced capabilities including a code interpreter, improved chat template, memory system, contextual search reformulation, and 128K context windows. Benchmarking shows clear improvements over previous GigaChat generations and competitive results against global leaders in coding, reasoning, and cross-domain tasks. Overall, GigaChat 3 Ultra empowers teams to explore frontier-scale AI without sacrificing transparency, customizability, or ecosystem compatibility.

GPT-4.1 mini is a streamlined version of GPT-4.1, offering the same core capabilities in coding, instruction adherence, and long-context comprehension, but with faster performance and lower costs. Ideal for developers seeking to integrate AI into real-time applications, GPT-4.1 mini maintains a 1 million token context window and is well-suited for tasks that demand low-latency responses. It is a cost-effective option for businesses that need powerful AI capabilities without the high overhead associated with larger models.

DeepSeekMath is an advanced 7B parameter language model created by DeepSeek-AI, specifically engineered to enhance mathematical reasoning capabilities within open-source language models. Building upon the foundation of DeepSeek-Coder-v1.5, this model undergoes additional pre-training utilizing 120 billion math-related tokens gathered from Common Crawl, complemented by data from natural language and coding sources. It has shown exceptional outcomes, achieving a score of 51.7% on the challenging MATH benchmark without relying on external tools or voting systems, positioning itself as a strong contender against models like Gemini-Ultra and GPT-4. The model's prowess is further bolstered by a carefully curated data selection pipeline and the implementation of Group Relative Policy Optimization (GRPO), which improves both its mathematical reasoning skills and efficiency in memory usage. DeepSeekMath is offered in various formats including base, instruct, and reinforcement learning (RL) versions, catering to both research and commercial interests, and is intended for individuals eager to delve into or leverage sophisticated mathematical problem-solving in the realm of artificial intelligence. Its versatility makes it a valuable resource for researchers and practitioners alike, driving innovation in AI-driven mathematics.

Introducing Mistral NeMo, our latest and most advanced small model yet, featuring a cutting-edge 12 billion parameters and an expansive context length of 128,000 tokens, all released under the Apache 2.0 license. Developed in partnership with NVIDIA, Mistral NeMo excels in reasoning, world knowledge, and coding proficiency within its category. Its architecture adheres to industry standards, making it user-friendly and a seamless alternative for systems currently utilizing Mistral 7B. To facilitate widespread adoption among researchers and businesses, we have made available both pre-trained base and instruction-tuned checkpoints under the same Apache license. Notably, Mistral NeMo incorporates quantization awareness, allowing for FP8 inference without compromising performance. The model is also tailored for diverse global applications, adept in function calling and boasting a substantial context window. When compared to Mistral 7B, Mistral NeMo significantly outperforms in understanding and executing detailed instructions, showcasing enhanced reasoning skills and the ability to manage complex multi-turn conversations. Moreover, its design positions it as a strong contender for multi-lingual tasks, ensuring versatility across various use cases.

GPT-4.1 represents a significant upgrade in generative AI, with notable advancements in coding, instruction adherence, and handling long contexts. This model supports up to 1 million tokens of context, allowing it to tackle complex, multi-step tasks across various domains. GPT-4.1 outperforms earlier models in key benchmarks, particularly in coding accuracy, and is designed to streamline workflows for developers and businesses by improving task completion speed and reliability.

DeepSeek-V2 is a cutting-edge Mixture-of-Experts (MoE) language model developed by DeepSeek-AI, noted for its cost-effective training and high-efficiency inference features. It boasts an impressive total of 236 billion parameters, with only 21 billion active for each token, and is capable of handling a context length of up to 128K tokens. The model utilizes advanced architectures such as Multi-head Latent Attention (MLA) to optimize inference by minimizing the Key-Value (KV) cache and DeepSeekMoE to enable economical training through sparse computations. Compared to its predecessor, DeepSeek 67B, this model shows remarkable improvements, achieving a 42.5% reduction in training expenses, a 93.3% decrease in KV cache size, and a 5.76-fold increase in generation throughput. Trained on an extensive corpus of 8.1 trillion tokens, DeepSeek-V2 demonstrates exceptional capabilities in language comprehension, programming, and reasoning tasks, positioning it as one of the leading open-source models available today. Its innovative approach not only elevates its performance but also sets new benchmarks within the field of artificial intelligence.

Llama 4 Scout is an advanced multimodal AI model with 17 billion active parameters, offering industry-leading performance with a 10 million token context length. This enables it to handle complex tasks like multi-document summarization and detailed code reasoning with impressive accuracy. Scout surpasses previous Llama models in both text and image understanding, making it an excellent choice for applications that require a combination of language processing and image analysis. Its powerful capabilities in long-context tasks and image-grounding applications set it apart from other models in its class, providing superior results for a wide range of industries.

Qwen3-Coder is an advanced code model that comes in various sizes, prominently featuring the 480B-parameter Mixture-of-Experts version (with 35B active) that inherently accommodates 256K-token contexts, which can be extended to 1M, and demonstrates cutting-edge performance in Agentic Coding, Browser-Use, and Tool-Use activities, rivaling Claude Sonnet 4. With a pre-training phase utilizing 7.5 trillion tokens (70% of which are code) and synthetic data refined through Qwen2.5-Coder, it enhances both coding skills and general capabilities, while its post-training phase leverages extensive execution-driven reinforcement learning across 20,000 parallel environments to excel in multi-turn software engineering challenges like SWE-Bench Verified without the need for test-time scaling. Additionally, the open-source Qwen Code CLI, derived from Gemini Code, allows for the deployment of Qwen3-Coder in agentic workflows through tailored prompts and function calling protocols, facilitating smooth integration with platforms such as Node.js and OpenAI SDKs. This combination of robust features and flexible accessibility positions Qwen3-Coder as an essential tool for developers seeking to optimize their coding tasks and workflows.

Phi-4-mini-reasoning is a transformer-based language model with 3.8 billion parameters, specifically designed to excel in mathematical reasoning and methodical problem-solving within environments that have limited computational capacity or latency constraints. Its optimization stems from fine-tuning with synthetic data produced by the DeepSeek-R1 model, striking a balance between efficiency and sophisticated reasoning capabilities. With training that encompasses over one million varied math problems, ranging in complexity from middle school to Ph.D. level, Phi-4-mini-reasoning demonstrates superior performance to its base model in generating lengthy sentences across multiple assessments and outshines larger counterparts such as OpenThinker-7B, Llama-3.2-3B-instruct, and DeepSeek-R1. Equipped with a 128K-token context window, it also facilitates function calling, which allows for seamless integration with various external tools and APIs. Moreover, Phi-4-mini-reasoning can be quantized through the Microsoft Olive or Apple MLX Framework, enabling its deployment on a variety of edge devices, including IoT gadgets, laptops, and smartphones. Its design not only enhances user accessibility but also expands the potential for innovative applications in mathematical fields.

Tülu 3 is a cutting-edge language model created by the Allen Institute for AI (Ai2) that aims to improve proficiency in fields like knowledge, reasoning, mathematics, coding, and safety. It is based on the Llama 3 Base and undergoes a detailed four-stage post-training regimen: careful prompt curation and synthesis, supervised fine-tuning on a wide array of prompts and completions, preference tuning utilizing both off- and on-policy data, and a unique reinforcement learning strategy that enhances targeted skills through measurable rewards. Notably, this open-source model sets itself apart by ensuring complete transparency, offering access to its training data, code, and evaluation tools, thus bridging the performance divide between open and proprietary fine-tuning techniques. Performance assessments reveal that Tülu 3 surpasses other models with comparable sizes, like Llama 3.1-Instruct and Qwen2.5-Instruct, across an array of benchmarks, highlighting its effectiveness. The continuous development of Tülu 3 signifies the commitment to advancing AI capabilities while promoting an open and accessible approach to technology.

Qwen3-Coder is a versatile coding model that comes in various sizes, prominently featuring the 480B-parameter Mixture-of-Experts version with 35B active parameters, which naturally accommodates 256K-token contexts that can be extended to 1M tokens. This model achieves impressive performance that rivals Claude Sonnet 4, having undergone pre-training on 7.5 trillion tokens, with 70% of that being code, and utilizing synthetic data refined through Qwen2.5-Coder to enhance both coding skills and overall capabilities. Furthermore, the model benefits from post-training techniques that leverage extensive, execution-guided reinforcement learning, which facilitates the generation of diverse test cases across 20,000 parallel environments, thereby excelling in multi-turn software engineering tasks such as SWE-Bench Verified without needing test-time scaling. In addition to the model itself, the open-source Qwen Code CLI, derived from Gemini Code, empowers users to deploy Qwen3-Coder in dynamic workflows with tailored prompts and function calling protocols, while also offering smooth integration with Node.js, OpenAI SDKs, and environment variables. This comprehensive ecosystem supports developers in optimizing their coding projects effectively and efficiently.

Z.ai has unveiled its latest flagship model, GLM-4.5, which boasts an impressive 355 billion total parameters (with 32 billion active) and is complemented by the GLM-4.5-Air variant, featuring 106 billion total parameters (12 billion active), designed to integrate sophisticated reasoning, coding, and agent-like functions into a single framework. This model can switch between a "thinking" mode for intricate, multi-step reasoning and tool usage and a "non-thinking" mode that facilitates rapid responses, accommodating a context length of up to 128K tokens and enabling native function invocation. Accessible through the Z.ai chat platform and API, and with open weights available on platforms like HuggingFace and ModelScope, GLM-4.5 is adept at processing a wide range of inputs for tasks such as general problem solving, common-sense reasoning, coding from the ground up or within existing frameworks, as well as managing comprehensive workflows like web browsing and slide generation. The architecture is underpinned by a Mixture-of-Experts design, featuring loss-free balance routing, grouped-query attention mechanisms, and an MTP layer that facilitates speculative decoding, ensuring it meets enterprise-level performance standards while remaining adaptable to various applications. As a result, GLM-4.5 sets a new benchmark for AI capabilities across numerous domains.

Claude Sonnet 4.6 represents a comprehensive upgrade to Anthropic’s Sonnet model line, delivering expanded capabilities across coding, reasoning, computer interaction, and professional knowledge tasks. With a beta 1M token context window, the model can process massive datasets such as full repositories, extended legal agreements, or multi-document research projects in a single request. Developers report improved reliability, better instruction adherence, and fewer hallucinations, making long working sessions smoother and more predictable. Early users preferred Sonnet 4.6 over its predecessor in the majority of tests and often selected it over Opus 4.5 for practical coding work. The model’s computer-use skills have advanced significantly, enabling it to navigate spreadsheets, complete web forms, and manage multi-tab workflows with near human-level competence in many cases. Benchmark evaluations show consistent performance gains across reasoning, coding, and long-horizon planning tasks. In competitive simulations like Vending-Bench Arena, Sonnet 4.6 demonstrated strategic capacity-building and profit optimization over time. On the developer platform, it supports adaptive and extended thinking modes, context compaction, and improved tool integration for greater efficiency. Claude’s API tools now automatically execute filtering and code-processing steps to enhance search and token optimization. Sonnet 4.6 is available across Claude.ai, Cowork, Claude Code, the API, and major cloud providers at the same starting price as Sonnet 4.5.

Kimi K2 Thinking is a sophisticated open-source reasoning model created by Moonshot AI, specifically tailored for intricate, multi-step workflows where it effectively combines chain-of-thought reasoning with tool utilization across numerous sequential tasks. Employing a cutting-edge mixture-of-experts architecture, the model encompasses a staggering total of 1 trillion parameters, although only around 32 billion parameters are utilized during each inference, which enhances efficiency while retaining significant capability. It boasts a context window that can accommodate up to 256,000 tokens, allowing it to process exceptionally long inputs and reasoning sequences without sacrificing coherence. Additionally, it features native INT4 quantization, which significantly cuts down inference latency and memory consumption without compromising performance. Designed with agentic workflows in mind, Kimi K2 Thinking is capable of autonomously invoking external tools, orchestrating sequential logic steps—often involving around 200-300 tool calls in a single chain—and ensuring consistent reasoning throughout the process. Its robust architecture makes it an ideal solution for complex reasoning tasks that require both depth and efficiency.

EverMemOS is an innovative memory-operating system designed to provide AI agents with a continuous and rich long-term memory, facilitating their ability to comprehend, reason, and develop over time. Unlike conventional “stateless” AI systems that forget previous interactions, this platform employs advanced techniques such as layered memory extraction, organized knowledge structures, and adaptive retrieval mechanisms to create coherent narratives from varied interactions. This capability allows the AI to reference past conversations, user histories, and stored information in a dynamic manner. On the LoCoMo benchmark, EverMemOS achieved an impressive reasoning accuracy of 92.3%, surpassing other similar memory-enhanced systems. Its core component, the EverMemModel, enhances parametric long-context understanding by utilizing the model’s KV cache, thus enabling a complete training process rather than depending solely on retrieval-augmented generation. This innovative approach not only improves the AI's performance but also ensures it can adapt to users' evolving needs over time.

GLM-4.7 Flash serves as a streamlined version of Z.ai's premier large language model, GLM-4.7, which excels in advanced coding, logical reasoning, and executing multi-step tasks with exceptional agentic capabilities and an extensive context window. This model, rooted in a mixture of experts (MoE) architecture, is fine-tuned for efficient inference, striking a balance between high performance and optimized resource utilization, thus making it suitable for deployment on local systems that require only moderate memory while still showcasing advanced reasoning, programming, and agent-like task handling. Building upon the advancements of its predecessor, GLM-4.7 brings forth enhanced capabilities in programming, reliable multi-step reasoning, context retention throughout interactions, and superior workflows for tool usage, while also accommodating lengthy context inputs, with support for up to approximately 200,000 tokens. The Flash variant successfully maintains many of these features within a more compact design, achieving competitive results on benchmarks for coding and reasoning tasks among similarly-sized models. Ultimately, this makes GLM-4.7 Flash an appealing choice for users seeking powerful language processing capabilities without the need for extensive computational resources.

This repository showcases the research preview of LongLLaMA, an advanced large language model that can manage extensive contexts of up to 256,000 tokens or potentially more. LongLLaMA is developed on the OpenLLaMA framework and has been fine-tuned utilizing the Focused Transformer (FoT) technique. The underlying code for LongLLaMA is derived from Code Llama. We are releasing a smaller 3B base variant of the LongLLaMA model, which is not instruction-tuned, under an open license (Apache 2.0), along with inference code that accommodates longer contexts available on Hugging Face. This model's weights can seamlessly replace LLaMA in existing systems designed for shorter contexts, specifically those handling up to 2048 tokens. Furthermore, we include evaluation results along with comparisons to the original OpenLLaMA models, thereby providing a comprehensive overview of LongLLaMA's capabilities in the realm of long-context processing.

GLM-4.1V is an advanced vision-language model that offers a robust and streamlined multimodal capability for reasoning and understanding across various forms of media, including images, text, and documents. The 9-billion-parameter version, known as GLM-4.1V-9B-Thinking, is developed on the foundation of GLM-4-9B and has been improved through a unique training approach that employs Reinforcement Learning with Curriculum Sampling (RLCS). This model accommodates a context window of 64k tokens and can process high-resolution inputs, supporting images up to 4K resolution with any aspect ratio, which allows it to tackle intricate tasks such as optical character recognition, image captioning, chart and document parsing, video analysis, scene comprehension, and GUI-agent workflows, including the interpretation of screenshots and recognition of UI elements. In benchmark tests conducted at the 10 B-parameter scale, GLM-4.1V-9B-Thinking demonstrated exceptional capabilities, achieving the highest performance on 23 out of 28 evaluated tasks. Its advancements signify a substantial leap forward in the integration of visual and textual data, setting a new standard for multimodal models in various applications.

Reka Flash 3 is a cutting-edge multimodal AI model with 21 billion parameters, crafted by Reka AI to perform exceptionally well in tasks such as general conversation, coding, following instructions, and executing functions. This model adeptly handles and analyzes a myriad of inputs, including text, images, video, and audio, providing a versatile and compact solution for a wide range of applications. Built from the ground up, Reka Flash 3 was trained on a rich array of datasets, encompassing both publicly available and synthetic information, and it underwent a meticulous instruction tuning process with high-quality selected data to fine-tune its capabilities. The final phase of its training involved employing reinforcement learning techniques, specifically using the REINFORCE Leave One-Out (RLOO) method, which combined both model-based and rule-based rewards to significantly improve its reasoning skills. With an impressive context length of 32,000 tokens, Reka Flash 3 competes effectively with proprietary models like OpenAI's o1-mini, making it an excellent choice for applications requiring low latency or on-device processing. The model operates at full precision with a memory requirement of 39GB (fp16), although it can be efficiently reduced to just 11GB through the use of 4-bit quantization, demonstrating its adaptability for various deployment scenarios. Overall, Reka Flash 3 represents a significant advancement in multimodal AI technology, capable of meeting diverse user needs across multiple platforms.

Sky-T1-32B-Preview is an innovative open-source reasoning model crafted by the NovaSky team at UC Berkeley's Sky Computing Lab. It delivers performance comparable to proprietary models such as o1-preview on various reasoning and coding assessments, while being developed at a cost of less than $450, highlighting the potential for budget-friendly, advanced reasoning abilities. Fine-tuned from Qwen2.5-32B-Instruct, the model utilized a meticulously curated dataset comprising 17,000 examples spanning multiple fields, such as mathematics and programming. The entire training process was completed in just 19 hours using eight H100 GPUs with DeepSpeed Zero-3 offloading technology. Every component of this initiative—including the data, code, and model weights—is entirely open-source, allowing both academic and open-source communities to not only replicate but also improve upon the model's capabilities. This accessibility fosters collaboration and innovation in the realm of artificial intelligence research and development.

The GPT-5.2 Thinking variant represents the pinnacle of capability within OpenAI's GPT-5.2 model series, designed specifically for in-depth reasoning and the execution of intricate tasks across various professional domains and extended contexts. Enhancements made to the core GPT-5.2 architecture focus on improving grounding, stability, and reasoning quality, allowing this version to dedicate additional computational resources and analytical effort to produce responses that are not only accurate but also well-structured and contextually enriched, especially in the face of complex workflows and multi-step analyses. Excelling in areas that demand continuous logical consistency, GPT-5.2 Thinking is particularly adept at detailed research synthesis, advanced coding and debugging, complex data interpretation, strategic planning, and high-level technical writing, showcasing a significant advantage over its simpler counterparts in assessments that evaluate professional expertise and deep understanding. This advanced model is an essential tool for professionals seeking to tackle sophisticated challenges with precision and expertise.

Devstral 2 represents a cutting-edge, open-source AI model designed specifically for software engineering, going beyond mere code suggestion to comprehend and manipulate entire codebases, which allows it to perform tasks such as multi-file modifications, bug corrections, refactoring, dependency management, and generating context-aware code. The Devstral 2 suite comprises a robust 123-billion-parameter model and a more compact 24-billion-parameter version, known as “Devstral Small 2,” providing teams with the adaptability they need; the larger variant is optimized for complex coding challenges that require a thorough understanding of context, while the smaller version is suitable for operation on less powerful hardware. With an impressive context window of up to 256 K tokens, Devstral 2 can analyze large repositories, monitor project histories, and ensure a coherent grasp of extensive files, which is particularly beneficial for tackling the complexities of real-world projects. The command-line interface (CLI) enhances the model's capabilities by keeping track of project metadata, Git statuses, and the directory structure, thereby enriching the context for the AI and rendering “vibe-coding” even more effective. This combination of advanced features positions Devstral 2 as a transformative tool in the software development landscape.

Qwen3 is a state-of-the-art large language model designed to revolutionize the way we interact with AI. Featuring both thinking and non-thinking modes, Qwen3 allows users to customize its response style, ensuring optimal performance for both complex reasoning tasks and quick inquiries. With the ability to support 119 languages, the model is suitable for international projects. The model's hybrid training approach, which involves over 36 trillion tokens, ensures accuracy across a variety of disciplines, from coding to STEM problems. Its integration with platforms such as Hugging Face, ModelScope, and Kaggle allows for easy adoption in both research and production environments. By enhancing multilingual support and incorporating advanced AI techniques, Qwen3 is designed to push the boundaries of AI-driven applications.

The GPT-5.2 Instant model represents a swift and efficient iteration within OpenAI's GPT-5.2 lineup, tailored for routine tasks and learning, showcasing notable advancements in responding to information-seeking inquiries, how-to guidance, technical documentation, and translation tasks compared to earlier models. This version builds upon the more engaging conversational style introduced in GPT-5.1 Instant, offering enhanced clarity in its explanations that prioritize essential details, thus facilitating quicker access to precise answers for users. With its enhanced speed and responsiveness, GPT-5.2 Instant is adept at performing common functions such as handling inquiries, creating summaries, supporting research efforts, and aiding in writing and editing tasks, while also integrating extensive enhancements from the broader GPT-5.2 series that improve reasoning abilities, manage longer contexts, and ensure factual accuracy. As a part of the GPT-5.2 family, it benefits from shared foundational improvements that elevate its overall reliability and performance for a diverse array of daily activities. Users can expect a more intuitive interaction experience and a significant reduction in the time spent searching for information.

Step 3.5 Flash is a cutting-edge open-source foundational language model designed for advanced reasoning and agent-like capabilities, optimized for efficiency; it utilizes a sparse Mixture of Experts (MoE) architecture that activates only approximately 11 billion of its nearly 196 billion parameters per token, ensuring high-density intelligence and quick responsiveness. The model features a 3-way Multi-Token Prediction (MTP-3) mechanism that allows it to generate hundreds of tokens per second, facilitating complex multi-step reasoning and task execution while efficiently managing long contexts through a hybrid sliding window attention method that minimizes computational demands across extensive datasets or codebases. Its performance on reasoning, coding, and agentic tasks is formidable, often matching or surpassing that of much larger proprietary models, and it incorporates a scalable reinforcement learning system that enables continuous self-enhancement. Moreover, this innovative approach positions Step 3.5 Flash as a significant player in the field of AI language models, showcasing its potential to revolutionize various applications.

Baichuan-13B is an advanced large-scale language model developed by Baichuan Intelligent, featuring 13 billion parameters and available for open-source and commercial use, building upon its predecessor Baichuan-7B. This model has set new records for performance among similarly sized models on esteemed Chinese and English evaluation metrics. The release includes two distinct pre-training variations: Baichuan-13B-Base and Baichuan-13B-Chat. By significantly increasing the parameter count to 13 billion, Baichuan-13B enhances its capabilities, training on 1.4 trillion tokens from a high-quality dataset, which surpasses LLaMA-13B's training data by 40%. It currently holds the distinction of being the model with the most extensive training data in the 13B category, providing robust support for both Chinese and English languages, utilizing ALiBi positional encoding, and accommodating a context window of 4096 tokens for improved comprehension and generation. This makes it a powerful tool for a variety of applications in natural language processing.