AI Language Model Library

Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks.

Meta's latest open model built for always-on local agents. 30B parameters, licensed under Apache 2.0 and runs on a single GPU — tuned for tool use, long tasks, and failure recovery.

This experimental preview of the architecture that will underpin Qwen4.

A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.

An efficient open embedding model featuring an extensive token context window.

The current, most capable model that runs on a single GPU.

Phi-4-mini brings significant enhancements in multilingual support, reasoning, and mathematics, and now, the long-awaited function calling feature is finally supported.

Cutting-edge large embedding model developed by mixedbread.ai.

Qwen3-Coder-Next is a coding-focused language model from Alibaba's Qwen team, optimized for agentic coding workflows and local development.

A strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token.

Phi-4 is a 14B parameter, state-of-the-art open model from Microsoft.

NVIDIA has tailored the Llama-3.1-Nemotron-70B-Instruct model to enhance the quality of responses generated by large language models. This customization aims to increase the usefulness of LLM outputs for user inquiries.

ShieldGemma consists of instructed models designed to assess the safety of text prompts and responses based on established safety guidelines.

Llama Guard 3 consists of a set of models specifically refined for classifying the content safety of inputs and responses from large language models (LLMs). These models are designed to enhance the reliability of content interactions.

Embedding models applied to extensive datasets at the sentence level.

BAAI's embedding model converts texts into vectors.

The BGE-M3, a recent model from BAAI, stands out for its versatility in multi-functionality, multilingual capabilities, and multi-granularity.

Snowflake offers a collection of performance-optimized text embedding models. These models are designed for enhanced efficiency.

Mistral Large 2 is the latest flagship model from Mistral, offering enhanced performance in code generation, mathematics, and reasoning, thanks to its 128k context window and support for numerous languages. Its capabilities are significantly improved compared to previous models.

Dolphin 3.0 Llama 3.1 8B is the next generation of the Dolphin series of instruct-tuned models designed to be the ultimate general purpose local model, enabling coding, math, agentic, function calling, and general use cases.

A compact reasoning model has been fine-tuned from the Qwen 2.5 3B Instruct model.

The IBM Granite 1B and 3B models are designed for low-latency applications and feature a long-context mixture of experts (MoE) architecture. These Granite models have been developed by IBM.

A group of effective AI models with fewer than 10 billion parameters excels in science, math, and coding thanks to advanced training methods. These models demonstrate high performance across these areas.

The IBM Granite Embedding 30M and 278M models models are text-only dense biencoder embedding models, with 30M available in English only and 278M serving multilingual use cases.

EXAONE 3.5 is a collection of instruction-tuned bilingual (English and Korean) generative models ranging from 2.4B to 32B parameters, developed and released by LG AI Research.

New state of the art 70B model. Llama 3.3 70B offers similar performance compared to the Llama 3.1 405B model.

Snowflake's frontier embedding model. Arctic Embed 2.0 adds multilingual support without sacrificing English performance or scalability.

Sailor2 are multilingual language models made for South-East Asia. Available in 1B, 8B, and 20B parameter sizes.

QwQ is the reasoning model of the Qwen series.

An open large reasoning model for real-world solutions by the Alibaba International Digital Commerce Group (AIDC-AI).

Tülu 3 is a leading instruction following model family, offering fully open-source data, code, and recipes by the The Allen Institute for AI.

Athene-V2 is a 72B parameter model which excels at code completion, mathematics, and log extraction tasks.

OpenCoder is an open and reproducible code LLM family which includes 1.5B and 8B models, supporting chat in English and Chinese languages.

Chirp Chirp! ? We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.

IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, thinking and structured JSON output. Released under Apache 2.0 license.

NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts (MoE) model with 3B active parameters built for always-on agents.

Our most capable model to date, designed for long-horizon work. 70.2% on Terminal-Bench 2.1 at 118B-A8B.

Laguna XS 2.1 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine.

A self-improving family of open-source models for agentic coding

North Mini Code is Cohere's first model for developers — a 30B Mixture-of-Experts model with 3B active parameters, built for agentic software engineering.

LFM2.5-8B-A1B, an edge model built for fast, reliable tool calling on consumer hardware.

A GPT-4o Level MLLM for Single Image, Multi Image and High-FPS Video Understanding on Your Phone

Granite Guardian 4.1 is a specialized safety and judging model from IBM Research that evaluates whether LLM prompts and responses meet specified harm criteria.

Mistral Medium 3.5 is the first flagship model of Mistral AI that merged instruction-following, reasoning, and coding in a single set of 128B weights.

IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, and structured JSON output. Released under Apache 2.0 license.

NVIDIA Nemotron 3 Nano Omni is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows.

DeepSeek-V4-Flash is the official release of DeepSeek-V4-Flash, built for efficient reasoning across a 1M-token context window, outperforming DeepSeek-V4-Pro (Preview).

DeepSeek-V4-Pro is a frontier Mixture-of-Experts model with a large context window and three reasoning modes.

Laguna XS.2 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine.

Qwen3.6 delivers substantial upgrades in agentic coding and thinking preservation than previous Qwen models.

MedGemma 1.5 4B is an updated version of the MedGemma 4B model.

MedGemma is a collection of Gemma 3 variants that are trained for performance on medical text and image comprehension.

An open 30B MoE model from NVIDIA with 3B activated parameters that delivers strong reasoning and agentic capabilities.

LFM2 is a family of hybrid models designed for on-device deployment. LFM2-24B-A2B is the largest model in the family, scaling the architecture to 24 billion parameters while keeping inference efficient.

NVIDIA Nemotron 3 Super is a 120B open MoE model activating just 12B parameters to deliver maximum compute efficiency and accuracy for complex multi-agent applications.

Qwen 3.5 is a family of open-source multimodal models that delivers exceptional utility and performance.

GLM-OCR is a multimodal OCR model for complex document understanding, built on the GLM-V encoder–decoder architecture.

LFM2.5 is a new family of hybrid models designed for on-device deployment.

As the strongest model in the 30B class, GLM-4.7-Flash offers a new option for lightweight deployment that balances performance and efficiency.

A new collection of open translation models built on Gemma 3, helping people communicate across 55 languages.

FunctionGemma is a specialized version of Google's Gemma 3 270M model fine-tuned explicitly for function calling.

Olmo is a series of Open language models designed to enable the science of language models. These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.

Nemotron-3-Nano is a new Standard for Efficient, Open, and Intelligent Agentic Models, now updated with a 4B parameter count model.

Olmo is a series of Open language models designed to enable the science of language models. These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.

EmbeddingGemma is a 300M parameter embedding model from Google.

Google Gemma 2 is now offered in three sizes: 2B, 9B and 27B.

DeepSeek-V3.1-Terminus is a hybrid model that supports both thinking mode and non-thinking mode.

Meta Llama 3 is the most advanced openly accessible language model available so far. It stands out as the leading choice among open-source LLMs.

OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases.

Qwen2 is a newly launched series of extensive language models developed by the Alibaba Group.

Alibaba's performant long context models for agentic and coding tasks.

A code language model based on an open-source Mixture-of-Experts approach, it delivers performance similar to GPT-4 Turbo for code-related tasks.

An update to Mistral Small that improves on function calling, instruction following, and less repetition errors.

Phi-3 consists of advanced, lightweight open models, including 3B (Mini) and 14B (Medium) versions, developed by Microsoft.

Gemma 3n models are designed for efficient execution on everyday devices such as laptops, tablets or phones.

Cohere has launched Aya 23, a cutting-edge family of multilingual models that accommodate 23 different languages.

Magistral is a small, efficient reasoning model with 24B parameters.

Mistral AI has released the 7B model, now updated to version 0.3.

Devstral: the best open source model for coding agents

Mistral AI has released a set of Mixture of Experts (MoE) models featuring open weights, available in parameter sizes of 8x7b and 8x22b.

Flagship vision-language model of Qwen and also a significant leap from the previous Qwen2-VL.

CodeGemma comprises robust, lightweight models capable of various coding tasks, including code completion, generation, natural language understanding, mathematical reasoning, and instruction adherence. These models are designed to efficiently handle diverse programming challenges.

Phi 4 reasoning and reasoning plus are 14-billion parameter open-weight reasoning models that rival much larger models on complex reasoning tasks.

Command R is a Large Language Model designed for effective conversational interaction and handling extended context tasks. It is optimized to facilitate engaging dialogue and manage lengthy discussions.

Phi 4 mini reasoning is a lightweight open model that balances efficiency with advanced reasoning ability.

Command R+ is a robust and scalable large language model designed specifically for optimal performance in retrieval augmented generation (RAG) and real-world enterprise applications.

Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.

LLaVA is a novel large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding. Updated to version 1.6.

IBM Granite 2B and 8B models are 128K context length language models that have been fine-tuned for improved reasoning and instruction-following capabilities.

Gemma is a cutting-edge family of lightweight open models developed by Google DeepMind. It has recently been updated to version 1.1.

DeepCoder is a fully open-Source 14B coder model at O3-mini level, with a 1.5B version also available.

Qwen 1.5 is a range of large language models developed by Alibaba Cloud, featuring parameter sizes from 0.5 billion to 110 billion.

Building upon Mistral Small 3, Mistral Small 3.1 (2503) adds state-of-the-art vision understanding and enhances long context capabilities up to 128k tokens without compromising text performance.

Llama 2 consists of a series of foundational language models with parameter sizes between 7 billion and 70 billion.

Cogito v1 Preview is a family of hybrid reasoning models by Deep Cogito that outperform the best available open models of the same size, including counterparts from LLaMA, DeepSeek, and Qwen across most standard benchmarks.

A powerful language model capable of generating and discussing code based on text prompts. It leverages natural language to facilitate coding tasks.

Meta's latest collection of multimodal models.

Eric Hartford developed the uncensored 8x7b and 8x22b fine-tuned models using the Mixtral mixture of experts, which are particularly effective for coding tasks.

EXAONE Deep exhibits superior capabilities in various reasoning tasks including math and coding benchmarks, ranging from 2.4B to 32B parameters developed and released by LG AI Research.

111 billion parameter model optimized for demanding enterprises that require fast, secure, and high-quality AI

The Uncensored Llama 2 model was developed by George Sung and Jarrad Hope.

DeepSeek Coder is a powerful coding model that has been trained on two trillion tokens of code and natural language. Its extensive training allows it to perform well in various coding tasks.

A new state-of-the-art version of the lightweight Command R7B model that excels in advanced Arabic language capabilities for enterprises in the Middle East and Northern Africa.

Phi-2 is a 2.7 billion parameter language model developed by Microsoft Research, showcasing exceptional reasoning and language comprehension skills.

The newest series of Code-Specific Qwen models features major enhancements in code generation, reasoning, and correction. These advancements lead to more effective coding solutions.

A compact and efficient vision-language model, specifically designed for visual document understanding, enabling automated content extraction from tables, charts, infographics, plots, diagrams, and more.

The uncensored Dolphin model, which is built on Mistral, performs exceptionally well in coding tasks. It has been updated to version 2.8.

Solar Pro Preview is an advanced large language model (LLM) featuring 22 billion parameters, optimized for deployment on a single GPU.

A versatile model with a parameter range of 3 billion to 70 billion, designed for use on entry-level hardware.

NVIDIA has developed a small language model that is optimized for commercial use, particularly in roleplay, retrieval-augmented generation (RAG) question answering, and function calling. It is designed to be user-friendly for various applications.

Granite-3.2 is a family of long-context AI models from IBM Granite fine-tuned for thinking capabilities.

Dolphin 2.9, developed by Eric Hartford and built on Llama 3, comes in 8B and 70B sizes. It boasts a range of capabilities in instruction, conversation, and coding.

The Qwen2.5 models have been pretrained on Alibaba's latest extensive dataset, which includes up to 18 trillion tokens. This model can handle up to 128K tokens and offers multilingual support.

A version of the DeepSeek-R1 model that has been post trained to provide unbiased, accurate, and factual information by Perplexity.

Bespoke Labs has created an advanced fact-checking model. This state-of-the-art technology is at the forefront of verifying information.

A fine-tuned version of Deepseek-R1-Distilled-Qwen-1.5B that surpasses the performance of OpenAI’s o1-preview with just 1.5B parameters on popular math evaluations.

Mistral OpenOrca is a 7 billion parameter model that has been fine-tuned from the Mistral 7B model with the OpenOrca dataset.

Mistral Small is an efficient, lightweight model tailored for affordable applications such as translation and summarization. Its design prioritizes performance while minimizing costs.

StarCoder2 represents the next evolution of openly trained LLMs for code, available in three sizes: 3B, 7B, and 15B parameters.

A collection of models designed to transform HTML content into Markdown format, facilitating content conversion tasks. These models are beneficial for efficiently handling content changes.

A fully open-source family of reasoning models built using a dataset derived by distilling DeepSeek-R1.

Zephyr comprises refined versions of the Mistral and Mixtral models, designed to serve as effective assistants. These models are specifically trained to enhance their helpfulness.

A collection of multimodal large language models (MLLMs) created for understanding the relationship between vision and language. These models are specifically tailored for tasks that involve interpreting both visual and textual information.

DeepSeek-R1 is a family of open reasoning models with performance approaching that of leading models, such as O3 and Gemini 2.5 Pro.

Yi 1.5 is an advanced bilingual language model that delivers exceptional performance. It excels in understanding and generating text in two languages.

An enhanced edition of DeekSeek-V2 that combines the general and coding capabilities of both DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct. This upgraded version integrates features from both platforms.

OLMo 2 is a new family of 7B and 13B models trained on up to 5T tokens. These models are on par with or better than equivalently sized fully open models, and competitive with open-weight models such as Llama 3.1 on English academic benchmarks.

A Llama 2-based model has been fine-tuned to enhance its ability to engage in Chinese dialogue. This adjustment aims to improve conversational performance in that language.

A highly effective model has been developed using a novel method known as Reflection-tuning, which enables a large language model to identify and rectify errors in its reasoning. This technique enhances the model's ability to improve its responses.

The smallest model in Cohere's R series delivers top-tier speed, efficiency, and quality to build powerful AI applications on commodity GPUs and edge devices.

A LLaVA model, fine-tuned from Llama 3 Instruct, has achieved improved scores across multiple benchmarks.

Yi-Coder is a collection of open-source programming language models that provide top-tier coding capabilities while using less than 10 billion parameters.

A general-purpose chat model built on Llama and Llama 2, offering context sizes ranging from 2K to 16K.

SmolLM2 is a series of compact language models offered in three sizes: 135M, 360M, and 1.7B parameters.

Nous Research offers a robust family of models that excel in scientific discussions and coding tasks. These powerful models are designed to enhance productivity in various research applications.

The IBM Granite Guardian 3.0 models 2B and 8B are created to identify risks in both prompts and responses. Their primary function is risk detection.

Wizard Vicuna Uncensored is a model with 7B, 13B, and 30B parameters, developed from Llama 2 by Eric Hartford. It is an uncensored version of the original model.

Cohere for AI has developed language models that excel in 23 diverse languages. These models are designed to deliver strong performance across all supported languages.

The TinyLlama SLM (small language model) is a compact AI language model with 1.1B parameters, designed for local on-premise inference on consumer grade hardware.

The IBM Granite 1B and 3B models are IBM's inaugural mixture of experts (MoE) Granite models, specifically created for low latency applications.

Codestral is the inaugural code model from Mistral AI, specifically developed for tasks involving code generation.

The IBM Granite 2B and 8B models are tailored for tool-based applications, facilitating retrieval augmented generation (RAG) to enhance code generation, translation, and bug fixing. These features help streamline various programming tasks.

StarCoder is a model designed for code generation, trained in over 80 programming languages.

Microsoft AI has developed a cutting-edge large language model that enhances performance in complex chat, multilingual interactions, reasoning, and agent applications.

An open-source model family trained on diverse data sets has outperformed ChatGPT on multiple benchmarks. It has been updated to version 3.5-0106.

Eric Hartford developed an experimental model with 1.1 billion parameters, utilizing the new Dolphin 2.8 dataset and based on TinyLlama.

OpenHermes 2.5 is a 7B model that has been fine-tuned by Teknium using Mistral and entirely open datasets.

Meta's Llama 3.2 introduces smaller models with 1 billion and 3 billion parameters. These compact versions aim to enhance performance and accessibility.

Advanced code generation model.

Stable Code 3B is a coding model that offers instruct and code completion options comparable to larger models like Code Llama 7B, which is 2.5 times its size.

CodeQwen1.5 is a substantial language model that has been pretrained using extensive code datasets.

A well-tuned Mistral model that effectively covers both domain and language.

The model specializes in solving mathematical and logical challenges. It emphasizes analytical reasoning and problem-solving skills.

Stable LM 2 is an advanced language model with 1.6 billion and 12 billion parameters, designed using multilingual data in several languages including English, Spanish, German, Italian, French, Portuguese, and Dutch.

A code generation model utilizing Code Llama.