Best AI Language Models Tools

Explore 75 AI language models tools to find the perfect solution.

Language Models

75 tools

Qwen2.5-7B

Qwen2.5-7B is a large language model designed for text generation, featuring improvements in coding, mathematics, instruction following, long text generation, and multilingual support. It supports context lengths up to 128K tokens and is intended for sophisticated NLP tasks.

DeepSeek-V3

An open-source language model from deepseek-ai offering efficient long-context inference with native sparse attention and support for multiple precisions.

Llama 3

Llama 3 is an open access large language model (LLM) released by Meta, available in various configurations (8B and 70B parameters) with capabilities for fine-tuning and integrations into platforms like Hugging Face, Google Cloud, and Amazon SageMaker.

UNfilteredAI-1B

A large-scale text generation model designed for creative and unconstrained content production without traditional filtering.

FLUX1.1 [pro]

A new text-to-image AI model capable of generating images six times faster than its predecessor, with higher quality, better prompt adherence, and more diversity in outputs. It includes a prompt upsampling feature that utilizes a language model to enhance prompts for improved image generation.

Shuttle-3

Shuttle-3 is a state-of-the-art language model designed for high-quality text generation, particularly suited for complex chat, multilingual communication, and reasoning tasks. It is fine-tuned from the Qwen-2.5-72b-Instruct model and designed to emulate high-quality prose similar to Claude 3 models.

WizardLM

WizardLM is a state-of-the-art large language model designed for complex chat, multilingual tasks, reasoning, and agent functionalities. It features an AI-powered pipeline (Auto Evol-Instruct) that optimizes instruction datasets for improved performance across various domains and leverages Arena Learning for an expanded learning pool of challenging instruction data.

Aria

A multimodal AI model that combines vision, language, and coding tasks, designed to deliver state-of-the-art performance across diverse tasks.

ToolACE-8B

ToolACE-8B is a finetuned LLaMA-3.1-8B-Instruct model designed for automatic tool usage and generating diverse tool-learning data, achieving state-of-the-art performance on the Berkeley Function-Calling Leaderboard. It features a novel self-evolution synthesis process and a dual-layer verification system for accurate data generation.

DeepSeek-R1

An autonomous agent designed for deep local and web research, capable of generating detailed reports with citations for various topics.

DeepSeek-V2

DeepSeek-V2 is a state-of-the-art Mixture-of-Experts (MoE) language model designed for economical training and efficient inference, boasting 236B total parameters with excellent performance across various benchmarks and exceptional capabilities in text generation and conversational AI.

DeepSeek-Coder-V2

An open‐source Mixture‐of‐Experts code language model that enhances code generation and reasoning capabilities for programming tasks. It supports an extended 128K context window and a wide array of programming languages, making it competitive with closed‐source models like GPT4-Turbo.

Watt Tool 8B

A tool built for enhancing AI workflow by enabling precise tool selection and multi-turn dialogue to support complex task automation.

watt-tool-70B

watt-tool-70B is a fine-tuned large language model based on LLaMa-3.3-70B-Instruct, optimized for advanced tool usage and multi-turn dialogue. It is designed for AI workflow building tasks, excelling in function calling and tool selection, and achieves state-of-the-art performance on the Berkeley Function-Calling Leaderboard (BFCL).

NSFW-gen-v2

An updated AI model that generates uncensored NSFW content, offering users the ability to produce imaginative and unfiltered outputs.

Bielik-11B-v2

Bielik-11B-v2 is an 11-billion parameter generative text model trained on Polish text corpora. Initialized from Mistral-7B-v0.2 and fine-tuned using advanced parallelization techniques, it offers robust text generation capabilities in Polish and English, as evidenced by its performance on multiple NLP leaderboards.

Marco-o1

An open-source large reasoning language model designed for complex real-world problems, leveraging chain-of-thought fine-tuning, Monte Carlo Tree Search, and self-reflection mechanisms to expand solution spaces and improve open-ended reasoning.

Llama-3.1-Tulu-3-8B

An instruction-following language model from AllenAI based on Llama 3.1, optimized for a wide range of NLP tasks including chat, math, and reasoning. It provides various fine-tuned versions (SFT, DPO, RLVR) along with extensive benchmarking and deployment guidance on Hugging Face.

Microsoft Phi-4

Microsoft Phi-4 is a state-of-the-art open language model (14B parameters, dense decoder-only transformer) trained on a blend of synthetic, public domain, and academic data. It has undergone rigorous supervised fine-tuning and direct preference optimization to improve instruction adherence, reasoning, and safety, making it suitable for research and generative AI applications.

Phi-3-mini-4k-instruct

A 3.8B parameter, lightweight instruction-tuned language model by Microsoft built on the Phi-3 datasets. It is designed for robust text generation, logical reasoning, and multi-turn conversation with support for both 4K and 128K token contexts.

MicroThinker-3B-Preview

MicroThinker-3B-Preview is a fine-tuned language model developed by huihui-ai, built from the Llama-3.2-3B-Instruct-abliterated base model. It is specifically optimized for enhanced reasoning capabilities and text generation, using the FineQwQ-142k dataset. The model card includes detailed training instructions and support for inference (e.g., via ollama).

DeepSeek-V3

A large AI model with a total of 685B parameters, supporting FP8 and BF16 modes and featuring multi-token prediction.

DeepSeek-R1

Open-weight reasoning LLM series with distilled variants; MIT-licensed.

Qwen

Qwen is the large language model family developed by Alibaba Cloud, showcased on Hugging Face. It includes advanced language and multimodal models (e.g., Qwen2.5, Qwen2.5-VL) that support text generation, image-text interactions, and long-context processing. The organization page also links to interactive demos like Qwen Chat, highlighting its practical application in AI chat and content generation.

Qwen/QwQ-32B-Preview

An experimental preview release large language model developed by the Qwen Team, featuring 32.5B parameters. It is designed to advance AI reasoning and text generation, supporting extended context lengths (up to 32,768 tokens) and built using transformer architectures with RoPE, SwiGLU, and RMSNorm. The model is geared towards research and demonstrates strong capabilities in math and coding, despite noted limitations in language consistency and common sense reasoning.

BLOOM

BLOOM is a multilingual large language model with 176 billion parameters developed by the BigScience project. It generates text in 46 natural languages and 13 programming languages, and is designed for research and deployment under a Responsible AI License. The release includes access to intermediary checkpoints, optimizer states, and is integrated into the Hugging Face ecosystem.

Open-r1

A fully open reproduction of DeepSeek-R1 that supports training with reasoning traces and scales across multiple nodes using TRL’s vLLM backend.

OpenAI GPT 1

OpenAI GPT 1 is the first transformer-based language model developed by OpenAI. It is a causal transformer pre-trained on a large corpus for language modeling and is available for inference through both PyTorch and TensorFlow. The model card provides comprehensive details including training methodology, risks, limitations, and usage guidelines.

GPT-2

GPT-2 is a pretrained generative transformer model by OpenAI, designed for text generation. It is trained using a causal language modeling objective on a large corpus of English text and is available on Hugging Face. The model card provides detailed usage examples, training procedure, limitations, and evaluation results.

OpenAI GPT-4o

OpenAI GPT-4o is an advanced multimodal AI model available via the Azure OpenAI Service. It integrates text, image, and audio processing to offer efficient and cost-effective performance, surpassing GPT-4 Turbo with Vision in speed, cost, and non-English language support. It is designed for enhanced customer service, advanced analytics, and content innovation.

OpenLLaMA

An open-source reproduction of Meta AI’s LLaMA large language model, offering 3B, 7B, and 13B parameter models trained on the RedPajama dataset with both PyTorch and JAX weights under the Apache-2.0 license.

Yi

A series of large language models, including variants like Yi-chat, acclaimed for performance in both English and Chinese on multiple benchmarks.

DeepSeek-Coder-V2-Lite-Instruct

An open-source Mixture-of-Experts code language model that provides advanced code intelligence, enabling functionalities comparable to GPT-4-Turbo for coding tasks.

DeepSeek-MoE

DeepSeek-MoE 16B is a Mixture-of-Experts (MoE) language model featuring 16.4B parameters. It employs fine-grained expert segmentation and shared experts isolation to achieve comparable performance to larger models with only around 40% of the typical computations. The repository includes both base and chat variants along with evaluation benchmarks and integration instructions via Hugging Face Transformers.

Perplexity R1-1776

A post-trained variant of the DeepSeek-R1 reasoning model by Perplexity AI, designed to remove censorship and deliver unbiased, accurate, and fact-based responses while maintaining robust reasoning skills.

DeepSeek-R1 Distill Qwen 14B GGUF

A quantized (GGUF) variant of the DeepSeek-R1 reasoning model distilled from Qwen 14B. This model supports a massive 128k context length and is tuned for reasoning and chain-of-thought tasks. It is provided by the lmstudio-community on Hugging Face, incorporating optimizations from llama.cpp.

Grok 3 ai

Grok 3 ai is xAI's flagship language model, introduced as an upgrade to Grok 2. It features enhanced computational power (10–15× more than its predecessor), advanced reasoning capabilities including a 'Big Brain Mode' for tackling complex multi-step problems, and a DeepSearch feature that scans and synthesizes information from the internet and social platforms. Additionally, it supports multimodal inputs and improved coding accuracy, positioning it as a strong competitor against models like GPT-4o and Gemini. The model is accessible via subscription plans integrated within X’s ecosystem.

EleutherAI/gpt-neox-20b

A 20-billion parameter autoregressive transformer language model developed by EleutherAI using the GPT-NeoX library. It is designed primarily for research purposes, with capabilities for further fine-tuning and adaptation, and provides detailed technical specifications and evaluation results.

Dolphin 3.0 R1 Mistral 24B

A next-generation instruct-tuned text generation model optimized for coding, math, reasoning, and agentic tasks. Built on the Mistral-24B base, it is fine-tuned with extensive reasoning traces to support function calling and steerable alignment, offering users local deployment control.

Falcon 3 Family

A family of open-source, decoder-only large language models under 10 billion parameters developed by Technology Innovation Institute (TII). The Falcon 3 models offer enhanced math, scientific, and coding capabilities through innovative pretraining techniques and are available in multiple variants including base and instruct configurations.

DeepSeek-R1-Distill-Qwen-1.5B

A distilled dense language model based on Qwen2.5-Math-1.5B that leverages the DeepSeek-R1 pipeline. It is designed for advanced reasoning, math, and code generation tasks, and is available under an MIT license with extensive evaluation metrics and deployment instructions on Hugging Face.

SmolLM

A family of efficient and compact AI models optimized for on-device usage in text and vision tasks.

FuseChat-7B-VaRM

FuseChat-7B-VaRM is a chat language model developed by FuseAI that fuses knowledge from multiple chat LLMs (NH2-Mixtral-8x7B, NH2-Solar-10.7B, and OpenChat-3.5-7B) using a fuse-then-merge strategy. It aims to integrate the strengths of diverse models into a single, memory-efficient LLM, achieving competitive performance on benchmarks like MT-Bench.

DeepSeek-R1-Distill-Qwen-14B

An open-source, distilled large language model derived from DeepSeek-R1 that is built on Qwen2.5-14B. It utilizes reinforcement learning techniques to achieve enhanced reasoning, chain-of-thought generation, and state-of-the-art performance benchmarks.

prunaai/hidream-l1-dev

An optimized version of the hidream-l1-dev model using the pruna ai optimisation toolkit. This model runs on Nvidia A100 GPUs, is available via an API on Replicate, supports rapid predictions (around 15 seconds per run), and has been executed over 28.5K times.

Llama 4 Maverick & Scout

A new generation of large language models from Meta released on Hugging Face. Llama 4 includes two Mixture-of-Experts models – Maverick (~400B total with 17B active parameters and 128 experts) and Scout (~109B total with 17B active parameters and 16 experts). Both support native multimodal inputs (text and images), extremely long context lengths (up to 10M tokens in Scout), and are integrated with Hugging Face transformers and TGI for easy deployment.

FinGPT

FinGPT is an open‐source repository that provides financial large language models along with training scripts, fine‐tuning techniques, and benchmark datasets. It is designed to efficiently adapt LLMs for financial applications, democratizing financial data and supporting research through released models and accompanying academic papers.

DeepSeek-V2-Lite

DeepSeek-V2-Lite is a Mixture-of-Experts language model designed for economical training and efficient inference. With 16B total parameters and 2.4B activated parameters, it employs innovative techniques such as Multi-head Latent Attention (MLA) and DeepSeekMoE for performance gains. The model is available for both text and chat completions via Hugging Face and is optimized to run with a 40GB GPU using BF16 precision.

Jamba-v0.1

Jamba-v0.1 is a state-of-the-art, hybrid SSM-Transformer large language model developed by AI21 Labs. It is a pretrained, mixture-of-experts generative text model with 12B active parameters (52B total across experts), supporting a 256K context length. Designed for high throughput, it serves as a strong base for fine-tuning into chat/instruct versions.

OpenAI GPT 4.1 API

OpenAI's flagship GPT-4.1 API is a high-performance large language model optimized for real-world applications. It supports up to 1M tokens of context, offers improved coding, advanced instruction following, enhanced formatting, and robust long-context comprehension, making it ideal for building intelligent agents, processing extensive documents, and handling complex workflows.

STILL-3-Tool-32B

A 32.8B parameter text-generation model that integrates Python code to enhance the reasoning process via tool manipulation. It achieves 81.70% accuracy on AIME 2024, matching o3-mini and outperforming o1 and DeepSeek-R1. The model is open-sourced on Hugging Face, and its design focuses on improving reasoning capabilities by leveraging integrated tool use.

spaCy Models

A GitHub repository by explosion that distributes pre-trained model packages for the spaCy NLP library. The repository provides model releases in .whl and .tar.gz formats for various NLP tasks (e.g., tagging, parsing, lemmatization, and named entity recognition) along with versioning and compatibility guidelines.

Llama4

Llama4 is a large autoregressive Mixture-of-Experts (MoE) multimodal model developed by Meta. It comes in two variants: Maverick (17B active parameters out of ~400B total with 128 experts) and Scout (17B active parameters out of ~109B total with 16 experts). The models support native multimodal inputs (text and images), long context lengths (up to 10 million tokens in some versions), and advanced quantization and offloading techniques for efficient deployment.

OpenAI GPT-4o API

GPT‑4o is OpenAI’s most advanced flagship multimodal model that supports text, image, and audio inputs and outputs, offering real-time responsiveness, a 1M token context window via API, and high performance across reasoning, math, and coding tasks. It is ideal for applications such as real-time voice assistants, interactive multimodal document Q&A, and advanced code generation.

Phi-4-mini-instruct

Phi-4-mini-instruct is a 3.8B-parameter lightweight language model from Microsoft, built from the Phi-4 family. It is designed for high-quality reasoning and instruction-following tasks, supports a 128K token context length, and is optimized via supervised fine-tuning and direct preference optimization. It is intended for both commercial and research use in memory/compute constrained and latency-sensitive environments.

Seed-Coder

Seed-Coder is a family of lightweight open‐source code language models (LLMs) that come in base, instruct, and reasoning variants (each around 8B parameters). Developed by ByteDance Seed, the models are designed to curate code training data automatically and enhance code generation and reasoning tasks.

deepfake-detector-model-v1

A deepfake detection image classification model fine-tuned from google/siglip2-base-patch16-512. It leverages the SiglipForImageClassification architecture to classify images as either 'fake' (deepfakes) or 'real', and is intended for applications such as media authentication, content moderation, forensic analysis, and security.

AM-Thinking-v1

AM-Thinking-v1 is a 32B dense language model built on Qwen 2.5-32B-Base, designed to enhance reasoning capabilities. It uses a post-training pipeline that includes supervised fine-tuning and dual-stage reinforcement learning, enabling strong performance in reasoning tasks like code generation, logic, and writing while operating efficiently on a single GPU.

Microsoft Phi-4-reasoning-plus

Phi-4-reasoning-plus is a state-of-the-art open-weight reasoning large language model developed by Microsoft. Finetuned from the base Phi-4 model with a mix of supervised fine-tuning on chain-of-thought traces and reinforcement learning, it is optimized for advanced reasoning tasks in math, science, and coding. The model features a dense 14B parameter decoder-only Transformer architecture with a 32k token context length and produces responses with a reasoning chain-of-thought followed by a summarization. It is intended for research and generative AI applications in constrained memory/latency settings.

Smaug-72B-v0.1

Smaug-72B-v0.1 is an open-source large language model for text generation developed by Abacus.AI. Based on Qwen-72B and finetuned using the novel DPO-Positive (DPOP) technique, it achieves high performance on benchmarks like MT-Bench and is the first open model to surpass an average score of 80% on the Open LLM Leaderboard.

DeepSeek-R1-Distill-Llama-8B

A distilled language model from the DeepSeek-R1 series built on the Llama-3.1-8B base. It is optimized for text generation and chain-of-thought reasoning tasks through reinforcement learning and selective fine-tuning, delivering competitive performance on math, code, and reasoning benchmarks.

Kimi-Dev

Kimi-Dev is an open-source coding LLM (Kimi-Dev-72B) designed for software engineering tasks such as automated code repair and test case generation. It uses large-scale reinforcement learning to autonomously patch repositories, ensuring that full test suites pass before accepting changes. The tool is available for download and deployment via GitHub and Hugging Face.

DeepCoder-14B-Preview

DeepCoder-14B-Preview is a code reasoning large language model fine-tuned from DeepSeek-R1-Distilled-Qwen-14B using distributed reinforcement learning (GRPO+ and iterative context lengthening) to enhance long-context code generation and reasoning. It achieves competitive scores on LiveCodeBench and other coding benchmarks, and is released under the MIT License on Hugging Face.

roberthein/modelname-new

This is an API reference page for the AI model 'modelname-new' by roberthein hosted on Replicate. The page shows usage statistics, a playground, API usage options, and links to documentation such as README and version history.

DeepSeek

DeepSeek is a verified Chinese company focused on advancing AGI by developing a suite of advanced text generation models (such as DeepSeek-R1 and DeepSeek-V3) available on Hugging Face. Their offerings include a range of models for text generation, research, and development, reflecting a robust commitment to AI model innovation.

Llama by Meta

Family of open-weight LLMs (text-only and vision) from Meta, including Llama 3.x and Llama 4 variants

MiniMax-M1

MiniMax-M1 is an open-weight, large-scale hybrid-attention reasoning model built using a hybrid Mixture-of-Experts architecture with a lightning attention mechanism. It supports an extended context length of up to 1 million tokens and is optimized with reinforcement learning for tasks ranging from mathematical reasoning to complex software engineering environments.

OpenAI GPT OSS

An open-source family of large language models released by OpenAI, featuring two models: a 117B-parameter model (gpt-oss-120b) and a 21B-parameter model (gpt-oss-20b). Utilizing mixture-of-experts (MoE) with 4-bit quantization (MXFP4), these models offer powerful reasoning, chain-of-thought capabilities, and tool use support, optimized for fast inference on modern GPUs, ranging from data-center H100s to consumer hardware.

OpenAI GPT-5

OpenAI GPT-5 is the latest language model from OpenAI, designed for advanced reasoning, code generation, instruction following, and tool usage. It is offered in several variants (gpt-5, gpt-5-mini, gpt-5-nano) to cater to tasks from complex multi-step reasoning to lightweight classification.

DeepSeek-V3.1-Base

DeepSeek-V3.1-Base is a highly advanced, long-context text generation model that supports both thinking and non-thinking modes. It introduces hybrid thinking mode, improved tool calling, and enhanced efficiency compared to preceding versions. With 671B parameters (37B activated) and a 128K context window, it is optimized using the UE8M0 FP8 scale format and represents a significant upgrade in tool usage and agent tasks. It is designed for complex conversational and code generation tasks.

Qwen3

Qwen3 is a series of large language models developed by the Qwen team at Alibaba Cloud. It offers multiple variants including Qwen3-Instruct-2507 and Qwen3-Thinking-2507 in different sizes, with advanced improvements in instruction following, reasoning, long-context understanding (up to 256K tokens, extendable to 1M), and various other capabilities for text generation and tool usage.

Granite-3.1-2B-Instruct

Granite-3.1-2B-Instruct is a lightweight, open-source 2B parameter instruct language model developed by the Granite Team and IBM, designed for instruction following tasks including summarization, problem-solving, text translation, reasoning, code tasks, function-calling, and more. It is optimized for long-context tasks and supports multilingual capabilities.

OpenAI GPT-5

OpenAI GPT-5 is OpenAI’s latest language model that excels at advanced reasoning, code generation, instruction following, and tool use. The webpage details its variants (gpt-5, gpt-5-mini, and gpt-5-nano), configurable reasoning levels (minimal, medium, high), verbosity settings, and integration best practices for coding, frontend development, and agentic tasks. It includes a playground, API integration, and pricing details.

Kimi-K2

Kimi-K2 is a state-of-the-art mixture-of-experts (MoE) large language model series developed by Moonshot AI. It features 1 trillion total parameters with 32 billion activated parameters, optimized for frontier knowledge, reasoning, coding tasks, and agentic capabilities including tool use and autonomous problem solving.

DeepSeek-V3.1

An AI tool optimized for smarter tool calling and enhanced agent task performance through post-training optimization techniques.