AI Intelligence Dashboard 2025
Interactive analysis of the global and Arab AI scene by AI Workshop
AI Models Database
Explore the leading models. Use filters to customize your search in over 200 models.
GPT-4o
OpenAI
A leading native multimodal model, integrating real-time text, audio, and image understanding for natural and fast human-like interaction.
GPT-4 Turbo
OpenAI
An enhanced version of GPT-4 with a massive context window (128k tokens) and knowledge updated to April 2023.
Claude 3.5 Sonnet
Anthropic
A leading model that balances intelligence and speed, excellent for coding, data analysis, and computer vision.
Claude 3 Opus
Anthropic
The most powerful model in the Claude 3 family, designed for highly complex tasks requiring near-human intelligence.
Claude 3 Haiku
Anthropic
The fastest and smallest model in the Claude 3 family, ideal for instant interactions and customer service.
Gemini 1.5 Pro
A multimodal model with a massive context window (1 million tokens), ideal for analyzing huge amounts of data.
Gemini 1.5 Flash
A lightweight and fast model from the Gemini 1.5 family, optimized for tasks requiring quick responses at scale.
Gemini 1.0 Ultra
The first model in the Gemini family, designed for highly complex tasks and excelling at understanding nuances.
Command R+
Cohere
A model designed for enterprise applications, focusing on reliable RAG, tool use, and multilingual performance.
Command R
Cohere
The base model of the Command family, balancing performance and cost for enterprise applications.
Mistral Large
Mistral AI
A leading language model from Mistral AI, featuring strong reasoning capabilities and high multilingual performance.
Grok-1.5
xAI
An improved version of Grok with enhanced reasoning capabilities and a longer context.
Jurassic-2 (J2)
AI21 Labs
A family of language models (Jumbo, Grande, Large) focusing on text quality and precise instruction following.
Inflection-2.5
Inflection AI
The model that powers the Pi assistant, designed for empathetic and helpful personal conversations.
Amazon Titan Text Premier
Amazon
Amazon's latest and largest language model, designed for advanced enterprise applications on the Bedrock platform.
Llama-3.1-405B
Meta
The most powerful open-source model, offering transparency and extensive customization at a large scale.
Llama-3.1-70B
Meta
A strong and balanced open-source model, providing excellent performance for a wide range of tasks.
Llama-3.1-8B
Meta
The smallest Llama 3.1 model, ideal for experiments and applications requiring high efficiency.
Mixtral 8x22B
Mistral AI
A high-performance open Mixture of Experts (MoE) model, using 8 experts for high efficiency.
Mixtral 8x7B
Mistral AI
The original MoE model that revolutionized open performance, fast and efficient.
Mistral 7B
Mistral AI
A small but very powerful model, considered an excellent starting point for many fine-tuned models.
Gemma 2 27B
The second generation of Gemma models, offering leading performance in its size class with high efficiency.
Gemma 2 9B
A mid-size version of Gemma 2, ideal for on-device applications.
Jais-30B-v3
Core42
The highest quality Arabic language model, open-source and bilingual (Arabic/English).
Aya 23
Cohere For AI
A multilingual model covering 23 languages, developed as part of a broad community effort.
Nemotron-4 340B
NVIDIA
A massive open-source model designed to generate synthetic data for training other models.
Phi-3 Family (SLM)
Microsoft
A family of open-source small language models (SLM) (3.8B, 7B, 14B), very efficient for on-device applications.
Qwen2
Alibaba Cloud
The second generation of Qwen models, with significant improvements in performance and multilingual capabilities.
DBRX
Databricks
A powerful open MoE model, showing excellent performance in coding and general tasks.
Falcon 180B
TII
One of the largest open models, developed in the UAE and known for its strong performance.
OLMo
AI2
A family of fully open-source models from the Allen Institute for AI, including training data and tools.
Yi-1.5
01.AI
The new generation of Yi models, available in various sizes and featuring strong conversation and vision capabilities.
OpenELM
Apple
A family of open and efficient language models from Apple, designed to work efficiently on-device.
Mamba
Carnegie Mellon & Princeton
A new model architecture (State Space Model) aiming to compete with the Transformer architecture with higher efficiency.
Vicuna
LMSYS
An open-source conversational model fine-tuned from Llama, known for its high quality in dialogue.
Zephyr
Hugging Face
A series of fine-tuned models that focus on better intent and instruction following.
BLOOM
BigScience
A massive multilingual language model, developed in an open collaborative effort involving over 1000 researchers.
MPT-30B
MosaicML
An open-source model from MosaicML (now part of Databricks) featuring a long context window and high quality.
Orca 2
Microsoft
A model that focuses on improving the reasoning abilities of small models by simulating the thought processes of larger models.
XLM-RoBERTa
Meta
A very powerful multilingual model, pre-trained on 2.5 terabytes of data in 100 languages.
WizardLM
Microsoft
A family of models fine-tuned using complex, self-generated instruction data, enhancing their reasoning capabilities.
OpenChat
OpenChat
A set of open-source conversational models fine-tuned with diverse data to achieve strong conversational performance.
Cerebras-GPT
Cerebras
A set of open-source language models released to demonstrate the capabilities of Cerebras hardware.
OpenHermes
Teknium
One of the most popular fine-tuned models from Mistral, known for its high quality in conversation and coding.
Dolphin
Eric Hartford
A family of uncensored fine-tuned models (usually from Llama or Mistral), aiming to provide direct answers.
SeaLLM
AI Singapore
A family of language models designed specifically for Southeast Asian languages and cultures.
BELLE
Lianjia
An open-source Chinese project focusing on producing high-quality instruction data and fine-tuned models.
ChatGLM3
Tsinghua University
The third generation of open bilingual (Chinese/English) conversational models.
InternLM2
Shanghai AI Lab
The second generation of InternLM models, with improved performance across various tasks.
LEIA
Large European AI consortium
A European consortium aiming to build large, open-source, multilingual language models specific to Europe.
GPT-SW3
AI Sweden
A set of large language models trained specifically for Nordic languages.
Nous-Hermes 2
Nous Research
One of the strongest fine-tuned models from Mistral and Llama, trained on high-quality synthetic data.
SOLAR-10.7B
Upstage
An open-source model that combines different models to achieve high performance while maintaining a relatively small size.
Stable Diffusion 3
Stability AI
A powerful open-source model for image generation, known for its flexibility and ability to understand complex text.
Stable Diffusion XL
Stability AI
An improved version of Stable Diffusion that provides higher image quality and better compositions.
Midjourney V6
Midjourney
A leading commercial model for generating artistic and realistic images, characterized by very high aesthetic quality.
DALL-E 3
OpenAI
The image generation model from OpenAI, integrated into ChatGPT and Bing, focusing on ease of use.
Imagen 3
Google's latest model for image generation, designed to create realistic and creative images with a deep understanding of details.
Ideogram 1.0
Ideogram AI
A model specialized in generating images that contain readable and clear text.
LLaVA-NeXT
various
A leading open-source vision-language model, combining large language models with image understanding capabilities for conversation about them.
Idefics2
Hugging Face
An open-source vision-language model from Hugging Face, known for its efficiency in handling images and text.
Florence-2
Microsoft
A unified and lightweight vision model that can handle a variety of tasks.
YOLOv10
Tsinghua University
Real-time object detection with high efficiency and excellent accuracy.
YOLOv8
Ultralytics
One of the most common and widely used versions of YOLO, known for its ease of use and strong performance.
YOLO-NAS
Deci AI
A new YOLO architecture found using Neural Architecture Search (NAS), to achieve an optimal balance between accuracy and speed.
ControlNet
Stanford University
A method for controlling image generation models by adding conditions such as poses or image edges.
CLIP
OpenAI
A foundational model that connects text and images, allowing for tasks like text-based image search and classification.
Segment Anything (SAM)
Meta
An image segmentation model that can identify and isolate any object in any image with a single click.
Vision Transformer (ViT)
A foundational model that showed the Transformer architecture can excel at computer vision tasks without needing CNNs.
DeepFloyd IF
Stability AI
An image generation model that relies on diffusion in pixel space, allowing for high quality and excellent text generation capability.
Kandinsky 3.0
Sber AI & others
The third generation of the image generation model, with improvements in quality and instruction-following ability.
StyleGAN3
NVIDIA
The third generation of StyleGAN, designed to create high-quality images with precise control over style.
ESRGAN
various
A popular model for image super-resolution, capable of adding realistic details.
DETR (DEtection TRansformer)
Meta
A model that treats object detection as a direct set prediction problem.
Swin Transformer
Microsoft Research
A hierarchical vision transformer architecture that achieves high efficiency by calculating attention within shifted windows.
DINOv2
Meta
A vision model that learns strong image representations through self-supervised learning, making it excellent for downstream tasks.
OpenPose
Carnegie Mellon University
A popular and influential library for real-time multi-person body, face, hand, and foot pose estimation.
SwinIR
various
An image restoration model based on Swin Transformer that achieves excellent results in tasks like denoising and super-resolution.
Mask R-CNN
Meta
A highly influential and foundational architecture for instance segmentation, which predicts a mask for each detected object.
U-Net
University of Freiburg
A convolutional neural network architecture originally developed for biomedical image segmentation, which has become a standard in the field.
CogVLM
Tsinghua University
A powerful open-source vision-language model, featuring excellent capabilities in complex visual conversation.
BLIP-2
Salesforce Research
An efficient architecture for training vision-language models, achieving excellent performance on tasks like captioning and visual question answering.
FastSAM
CASIA
A fast alternative to the original SAM model, achieving similar results at a much higher speed, making it suitable for real-time applications.
Sora
OpenAI
An advanced text-to-video model, capable of creating realistic and complex scenes up to a minute long.
KLING
Kuaishou
A competing Chinese model for generating realistic, high-definition (1080p) videos up to two minutes long.
Lumiere
A text-to-video model that uses a Space-Time U-Net architecture to generate realistic and coherent motion.
Pika 1.0
Pika Labs
An easy-to-use platform for generating and editing video with AI.
Runway Gen-2
RunwayML
One of the leading models in the market for generating video from text or images.
Stable Video Diffusion
Stability AI
An open-source model for generating short video clips from images.
Make-A-Video
Meta
One of the early and influential models in the text-to-video space.
VEO
Google's latest model for generating high-quality (1080p+) video with a deep understanding of language and cinematic concepts.
Whisper-large-v3
OpenAI
An accurate speech-to-text model, supporting multiple languages with great proficiency and efficiency.
Suno V3
Suno AI
An advanced model for generating full songs and music (including vocals) from a simple text description.
Udio
Udio AI
A powerful model for generating high-quality music and songs, considered a major competitor to Suno.
MusicGen
Meta
A model for generating high-quality music clips from a text description.
XTTS-v2
Coqui AI
A high-quality text-to-speech model with voice cloning across multiple languages.
ElevenLabs TTS
ElevenLabs
A leading platform for realistic speech generation and voice cloning, known for its extremely natural sound quality.
Bark
Suno AI
An open-source text-to-speech model, capable of generating very realistic voices with non-verbal cues.
VALL-E X
Microsoft
A text-to-speech model capable of cloning a person's voice from a short sample (3 seconds).
SeamlessM4T v2
Meta
A comprehensive translation model supporting speech-to-speech, speech-to-text, and text-to-speech across 100 languages.
Wav2Vec 2.0
Meta
A foundational model for learning speech representations from unlabeled data, widely used in speech recognition.
Tacotron 2
A neural network architecture for generating speech directly from text, which is the basis for many modern TTS systems.
Riffusion
Riffusion
A model that uses Stable Diffusion to generate spectrograms and then converts them into music.
Jukebox
OpenAI
An early and powerful model for generating music with rudimentary vocals, known for its high quality but very slow speed.
MMS (Massively Multilingual Speech)
Meta
A massive project providing models for speech recognition, generation, and language identification for over 1100 languages.
VITS
various
A popular speech generation architecture that combines VAEs and GANs to efficiently generate high-quality speech.
CLAP
Microsoft
A model that connects text and audio, allowing for tasks like text-based audio search and classification.
StyleTTS 2
various
A text-to-speech model that achieves very high quality with precise style control without needing a reference voice.
AlphaCode 2
Google DeepMind
An advanced code generation model that solves competitive programming problems at a level comparable to top programmers.
GitHub Copilot
GitHub (Microsoft)
The most popular AI-powered programming assistant, using OpenAI models to provide smart code suggestions.
Code Llama 70B
Meta
A model specialized in generating and understanding code, one of the most powerful open models in this field.
DeepSeek-Coder-V2
DeepSeek AI
A very powerful open-source model specialized in programming, supporting 338 programming languages.
CodeGemma
A family of open and lightweight models specialized in efficient code completion and generation.
StarCoder 2
BigCode (ServiceNow & Hugging Face)
The second generation of StarCoder models, trained on a massive amount of code.
CodeGen2
Salesforce
An open-source model for code generation, capable of working with multiple programming languages.
WizardCoder
WizardLM Team
A Llama model fine-tuned on a large programming dataset, achieving excellent performance.
Phind-Coder
Phind
An open-source programming model fine-tuned on high-quality data, known for its speed and accuracy.
AlphaFold 3
Google DeepMind
A revolutionary model that predicts the structure of proteins and molecular interactions with ultra-high accuracy.
ESMFold
Meta
A faster alternative to AlphaFold for protein structure prediction, based on language models.
GraphCast
Google DeepMind
An AI model for weather forecasting that surpasses traditional methods in accuracy and speed.
Med-PaLM 2
A language model specialized in the medical field, for answering medical questions and summarizing health data.
BioGPT
Microsoft
A large language model pre-trained on biomedical literature for information extraction.
Galactica
Meta
A language model trained on scientific papers (was withdrawn but influential).
RoseTTAFold
University of Washington
Another powerful model for protein structure prediction, considered a significant competitor to AlphaFold.
FourCastNet
NVIDIA
A weather forecasting model based on computer vision, known for its extreme speed.
AlphaTensor
Google DeepMind
A reinforcement learning model that discovered new and faster algorithms for matrix multiplication, a fundamental computation.
GNoME
Google DeepMind
A model that discovered 2.2 million new crystal materials, significantly accelerating the process of materials discovery.
ProGen
Salesforce Research
A language model for generating proteins with specific functions by learning from protein sequences.
MolGPT
various
A modified language model for generating new molecules with desired chemical properties.
TripoSR
Stability AI & Tripo AI
A fast, open-source model for generating 3D models from a single image.
Shap-E
OpenAI
A model for generating 3D models from text or images, producing implicit representations.
Point-E
OpenAI
A model for generating 3D models from text, focusing on creating a point cloud.
DreamFusion
A model that uses a 2D image generation model to generate coherent 3D scenes from text.
NeRF (Neural Radiance Fields)
UC Berkeley & Google
A foundational technique for generating novel views of complex 3D scenes from a small set of images.
GET3D
NVIDIA
A model that generates high-quality 3D meshes with rich details and textures.
Gaussian Splatting
INRIA
A new technique for representing and rendering 3D scenes in real-time with high quality, considered a fast alternative to NeRF.
AlphaGo
Google DeepMind
The historic model that defeated the world champion in the game of Go, marking a milestone in AI.
MuZero
Google DeepMind
An advanced model that learns the rules of the game itself and plans to achieve superhuman performance.
RT-2 (Robotics Transformer 2)
Google DeepMind
A vision-language-action model that transfers knowledge from the internet to robot control.
AlphaStar
Google DeepMind
A model that reached Grandmaster level in the strategically complex game of StarCraft II.
DreamerV3
Google DeepMind
A general reinforcement learning agent that can master a wide range of tasks with a fixed compute resource.
Gato
Google DeepMind
A single general agent that can perform over 600 different tasks, including playing, chatting, and controlling a robotic arm.
PPO (Proximal Policy Optimization)
OpenAI
A very popular reinforcement learning algorithm, used to train many agents due to its stability and ease of implementation.
PaLM-E
A multimodal model that integrates vision and language into a single model for robot control.
SayCan
A model that connects language models with robotic capabilities, allowing robots to understand and execute high-level instructions.
AlphaDev
Google DeepMind
A reinforcement learning model that discovered faster sorting algorithms, demonstrating its ability to improve fundamental code.
DQN (Deep Q-Network)
Google DeepMind
The first deep reinforcement learning algorithm to achieve human-level performance on Atari games, a milestone in the field.
DeepAR
Amazon
A popular time series forecasting model that uses recurrent neural networks (RNN) to produce probabilistic forecasts.
N-BEATS
Element AI, MILA
A deep learning architecture for time series forecasting that achieves high performance without needing prior domain knowledge.
Temporal Fusion Transformer (TFT)
A transformer-based model for multi-horizon time series forecasting, combining different types of inputs.
LightGCN
National University of Singapore
A simplified and effective graph neural network model for recommendation, focusing on the neighborhood in the user-item graph.
TimesNet
Tsinghua University
A new model for time series forecasting that discovers multiple periodic patterns in time data.
DCN V2 (Deep & Cross Network)
Google & Stanford
A popular recommendation model that combines deep networks and cross networks to effectively capture feature interactions.
TAPAS
Google AI
A BERT model modified to answer questions over tables, allowing for understanding of tabular data.
ERNIE
Baidu
A series of language models from Baidu that focus on integrating knowledge from knowledge graphs.
PanGu-Σ
Huawei
A large language model from Huawei with strong performance in Chinese and general tasks.
HyperCLOVA X
Naver
A large language model from Naver (South Korea) designed for wide-scale applications.
GAN (Generative Adversarial Network)
University of Montreal
A foundational framework consisting of two networks (generator and discriminator) that compete to create realistic data, the basis for many generative models.
VAE (Variational Autoencoder)
University of Amsterdam
A generative model that learns a latent representation of data, widely used in tasks like image generation and dimensionality reduction.
TabNet
Google Research
A deep learning model for tabular data that uses sequential attention to select important features at each decision step.
XGBoost
DMLC
An optimized, distributed, and high-performance gradient boosting library. Widely used in competitions and production.
LightGBM
Microsoft
A gradient boosting framework that uses histogram-based techniques for high speed and memory efficiency.
CatBoost
Yandex
A gradient boosting algorithm that handles categorical features automatically and efficiently.
Scikit-learn
Community
The essential Python library for machine learning, providing simple and effective tools for data mining and analysis.
Prophet
Meta
A library for forecasting time series data, designed to handle data with strong seasonal trends and holidays.
Isolation Forest
various
An effective algorithm for anomaly detection by isolating them instead of identifying normal regions.
SVM (Support Vector Machine)
various
A powerful classification algorithm that finds the best hyperplane that separates two classes of data.
BERT
A foundational model that revolutionized natural language understanding through bidirectional attention.
T5 (Text-To-Text Transfer Transformer)
A model that treats all NLP tasks as a "text-to-text" problem.
RoBERTa
Meta
A robust optimization of the BERT model, trained longer on more data.
GPT-2
OpenAI
A model that showed surprising abilities in generating coherent text, its initial release was controversial.
Transformer
The "Attention Is All You Need" paper that introduced the Transformer architecture, the basis for most modern models.
Word2Vec
A foundational technique for creating "word embeddings," which capture semantic relationships between words.
AlexNet
University of Toronto
A convolutional neural network that won the 2012 ImageNet competition, sparking the modern deep learning revolution.
ResNet
Microsoft Research
A deep neural network architecture that introduced "residual connections" to train much deeper networks.
EfficientNet
A family of models that uses a compound method to efficiently scale networks (depth, width, resolution).
Stanford Alpaca
Stanford University
An influential project that showed a 7B Llama model could be fine-tuned on little data to achieve high performance.
Dolly 2.0
Databricks
The first open-source instruction-following model trained on a human-generated dataset licensed for commercial use.
Guanaco
University of Washington
An improved version of Alpaca that uses the QLoRA technique to allow fine-tuning of large models on a single GPU.
Play.ht 2.0
Play.ht
An advanced commercial platform for generating very realistic voices and high-quality voice cloning.
OPT (Open Pre-trained Transformer)
Meta
A series of open-source language models, released to help promote transparency and research in large models.