Explore AI Models

OpenAI's most advanced multimodal model. Accepts text, image, and audio inputs with best-in-class reasoning. ~200B parameters (estimated).

Latest image generation model with improved prompt understanding and photorealistic outputs.

Open-source speech recognition model. Robust multilingual transcription. 1.55B parameters.

Google's fastest model with multimodal reasoning across text, image, and audio.

Anthropic's most intelligent model. Excels at complex reasoning, coding, and analysis.

Meta's largest open-source model. State-of-the-art with 128K context. 405B parameters.

Previous flagship model for reasoning and code generation. 70B parameters (estimated).

671B MoE model with 37B active parameters. Rivals frontier models at fraction of cost.

xAI's frontier model with real-time information access and strong reasoning.

Scalable enterprise model optimized for RAG, tool use, and multi-step agents. 104B parameters.

Popular open-source image generation model with photorealistic output. 3.5B parameters.

Flagship model with strong bilingual English-Chinese capabilities.

Strong bilingual model excelling at coding and reasoning. 34B parameters.

Compact bilingual model for diverse applications. 9B parameters.

Specialized code model with strong performance for its size. 9B parameters.

Alternative architecture model using hyena operator. 7B parameters.

Enterprise-grade MoE model optimized for efficiency. 132B total / 36B active parameters.

Instruction-tuned DBRX for enterprise applications. 132B total parameters.

Apple's efficient open-source language model family. 3B parameters.

Advanced video generation model with high fidelity and temporal coherence.

Text-to-video model with scene-level physics understanding.

Chinese video generation model with strong motion and physics simulation.

Open-source text-to-video model with strong motion quality. 10B parameters.

Fast open-source video generation model based on DiT architecture. 2B parameters.

Industry-leading image generation with exceptional artistic quality and photorealism.

Commercial-safe image generation trained on licensed content.

Image generation model with industry-best text rendering in images.

Image generation model optimized for graphic design and aesthetics.

Compact and cost-efficient variant of GPT-4o. Strong performance on everyday tasks. ~8B parameters (estimated).

High-performance GPT-4 variant with 128K context window and improved instruction following. ~1.8T parameters (estimated).

OpenAI's flagship large language model with advanced reasoning. ~1.8T parameters (estimated MoE).

Fast, affordable model optimized for chat and instruction-following. 175B parameters.

OpenAI's reasoning model with chain-of-thought thinking for complex problem-solving. ~200B parameters (estimated).

Smaller reasoning model optimized for STEM and coding tasks. ~100B parameters (estimated).

Enhanced reasoning model with improved reliability for the most demanding tasks.

Next-generation reasoning model with state-of-the-art benchmark performance.

Compact reasoning model balancing speed, cost, and strong reasoning.

Latest compact reasoning model with enhanced multimodal and agentic capabilities.

Faster variant of Whisper v3 with pruned decoder. 809M parameters.

Mid-size speech recognition model balancing accuracy and speed. 769M parameters.

Compact speech recognition model for resource-constrained environments. 244M parameters.

Text-to-video generation model capable of creating realistic and imaginative scenes.

Code-focused model powering GitHub Copilot. Excels at code generation. 12B parameters.

Incremental improvement over GPT-4 with better coding performance and instruction following.

Compact variant of GPT-4.1 optimized for cost-effective deployments.

Ultra-compact GPT-4.1 variant designed for the fastest, cheapest inference at scale.

Google's most advanced thinking model with enhanced reasoning and 1M token context.

Fast, efficient Gemini model with thinking capabilities and multimodal understanding.

Powerful multimodal model with 2M token context window for long-document understanding.

Lightweight model optimized for speed and cost-efficiency with strong multimodal abilities.

Google's first Gemini flagship model for highly complex multimodal tasks.

Google's latest open-source model. State-of-the-art for its size class. 27B parameters.

Mid-size open-source Gemma 3 model balancing capability and efficiency. 12B parameters.

Compact open-source Gemma 3 model for mobile and edge deployment. 4B parameters.

Ultra-compact Gemma 3 model for on-device inference. 1B parameters.

Open-source text model with strong general capabilities. 27B parameters.

Efficient open-source model for balanced performance. 9B parameters.

Compact open-source model ideal for edge and mobile inference. 2B parameters.

Specialized code model based on Gemma for code generation and completion. 7B parameters.

Google's advanced language model powering Bard and other AI features. 340B parameters (estimated).

Highest-quality text-to-image model with photorealistic output and deep language understanding.

Most capable video generation model producing high-quality, realistic footage.

Text-to-text transformer for NLP tasks. Pre-trained on C4 corpus. 11B parameters.

Instruction-tuned T5 model with strong zero-shot capabilities. 11B parameters.

Bidirectional encoder for NLU tasks. Foundation of modern NLP. 340M parameters.

Vision-language model for zero-shot image classification using sigmoid loss. 878M parameters.

Anthropic's most powerful model with unmatched reasoning, analysis, and creative capabilities.

Balanced Claude 4 model offering strong performance with improved speed and efficiency.

Hybrid thinking model combining instant responses with extended reasoning when needed.

Fast, affordable Claude model for quick responses and high-throughput applications.

Previous flagship with exceptional performance on highly complex, open-ended tasks.

Balanced model with strong performance and practical speed for enterprise workloads.

Fastest Claude 3 model for near-instant responses at scale.

Meta's latest MoE model with 17B active / 109B total parameters and 10M token context.

Larger Llama 4 MoE variant: 17B active / 400B total parameters for complex reasoning.

Cost-effective 70B model delivering Llama 3.1 405B-level performance. 70B parameters.

High-performance open-source model for commercial applications. 70B parameters.

Compact Llama model for edge deployment and constrained environments. 8B parameters.

Previous generation large Llama model with strong general performance. 70B parameters.

Efficient previous-gen Llama model for everyday applications. 8B parameters.

Large open-source model that kickstarted the open LLM revolution. 70B parameters.

Mid-size Llama 2 model balancing capability and compute. 13B parameters.

Compact foundational model for fine-tuning and experimentation. 7B parameters.

Largest Code Llama variant for complex code generation. 70B parameters.

Specialized coding model supporting code generation and infilling. 34B parameters.

Compact code model for fast code completion and generation. 7B parameters.

Safety classifier for content moderation in AI applications. 8B parameters.

Foundation model for promptable visual segmentation in images and videos.

Music generation model creating audio from text descriptions. 3.3B parameters.

Mid-size music generation model for text-to-music. 1.5B parameters.

Multilingual multimodal translation model for speech and text. 2.3B parameters.

Embedding model binding 6 modalities: images, text, audio, depth, thermal, IMU.

Mistral's flagship model for complex reasoning and multilingual tasks. 123B parameters.

Balanced model for business applications at moderate cost.

24B parameter open-source model competitive with much larger alternatives.

Efficient model for simple tasks, very low latency, and bulk processing.

Open-source model co-developed with NVIDIA. 128K context. 12B parameters.

The original Mistral model that outperformed Llama 2 13B. 7.3B parameters.

Large MoE model. 141B total / 39B active parameters. Strong multilingual.

Efficient MoE model. 46.7B total / 12.9B active parameters.

Latest specialized coding model for code generation and review. 22B parameters.

Linear-time coding model using Mamba architecture. 7B parameters.

Frontier multimodal model for document understanding and image analysis. 124B parameters.

Open-source multimodal model with vision capabilities. 12B parameters.

Reasoning model using RL to achieve o1-level performance. 671B total / 37B active parameters.

R1 reasoning distilled into Qwen 2.5 32B architecture. 32B parameters.

R1 reasoning distilled into Llama 3.1 70B architecture. 70B parameters.

R1 reasoning distilled into compact 7B architecture. 7B parameters.

Ultra-compact R1 distillation for edge deployment. 1.5B parameters.

Previous generation MoE model with strong general capabilities. 236B total parameters.

Code model supporting 338 programming languages. 236B total / 21B active parameters.

Specialized code generation model trained from scratch. 33B parameters.

Compact code model for fast development assistance. 6.7B parameters.

Multimodal model unifying visual understanding and generation. 7B parameters.

xAI's most powerful model topping multiple benchmarks. Real-time information access.

Lightweight thinking model with controllable reasoning depth.

Compact Grok variant for fast, cost-effective inference.

xAI's first open-source model released as MoE. 314B total parameters.

Latest enterprise model with 256K context and multilingual agentic performance. 111B parameters.

Efficient enterprise model for production RAG workloads. 35B parameters.

Compact open-source model for fast enterprise inference. 7B parameters.

Latest embedding model for semantic search, classification, and clustering.

Multilingual embedding model supporting 100+ languages.

Open-source multilingual model covering 23 languages. 35B parameters.

Compact multilingual model for 23 languages. 8B parameters.

Instruction-tuned multilingual model excelling in 23 languages. 32B parameters.

Latest Stable Diffusion with improved text rendering and composition. 8B parameters.

Mid-size diffusion model balancing quality and speed. 2.5B parameters.

Classic and widely-used image generation model with huge ecosystem. 860M parameters.

Open-source video generation model for short clips from images. 1.5B parameters.

AI music and sound generation from text descriptions. 1.1B parameters.

Open-source language model for general-purpose text generation. 12B parameters.

Small language model excelling at reasoning and STEM tasks. 14B parameters.

Mixture-of-experts variant of Phi-3.5 for improved efficiency. 42B total / 6.6B active.

Compact model with 128K context for edge/mobile deployment. 3.8B parameters.

Mid-size Phi-3 model with strong reasoning capabilities. 14B parameters.

Compact language model for on-device AI applications. 3.8B parameters.

Efficient small model demonstrating emergent capabilities. 2.7B parameters.

Unified vision model for captioning, detection, segmentation, and OCR. 770M parameters.

Foundation vision model for diverse visual understanding tasks. 230M parameters.

Research model trained with improved reasoning strategies. 13B parameters.

MoE model with enhanced instruction following. 141B total parameters.

Highly capable multimodal model balancing accuracy, speed, and cost.

Fast, low-cost multimodal model processing images, video, and text.

Text-only model with lowest latency at very low cost.

Image generation model for creative and professional content creation.

Video generation model for short-form content creation.

Amazon's flagship text model for enterprise workloads.

Text embedding model for search and retrieval applications.

Largest Qwen 3 MoE model with hybrid thinking. 235B total / 22B active parameters.

Strong general-purpose model with hybrid thinking. 32B parameters.

Mid-size Qwen 3 model balancing power and efficiency. 14B parameters.

Compact Qwen 3 model for versatile applications. 8B parameters.

Small Qwen 3 model for edge deployment. 4B parameters.

Ultra-compact Qwen 3 model for mobile devices. 1.7B parameters.

Tiny Qwen 3 model for embedded and IoT applications. 0.6B parameters.

Large model competitive with Llama 3.1 405B on reasoning benchmarks. 72B parameters.

Strong mid-size model for diverse tasks. 32B parameters.

Efficient model for production deployments. 14B parameters.

Compact model for fast inference and fine-tuning. 7B parameters.

Small model for resource-constrained environments. 3B parameters.

Top open-source coding model rivaling GPT-4o on coding benchmarks. 32B parameters.

Compact code model for fast code completion. 7B parameters.

Flagship vision-language model for document/chart understanding and visual QA.

Balanced vision-language model for image-based tasks.

Reasoning model by Alibaba matching o1-mini level performance. 32B parameters.

Fast and powerful model rivaling frontier models at a fraction of cost.

Open-source text-to-image model with strong Chinese text support. 8B parameters.

Open-source DiT-based image generation with 4K resolution support. 900M parameters.

LLM-based text embedding model with strong retrieval performance. 7B parameters.

Multi-lingual, multi-granularity embedding model. 568M parameters.

Text embedding model based on Qwen2 for retrieval tasks. 7B parameters.

State-of-the-art generalist embedding model. 7B parameters.

Multilingual multi-task embedding model with matryoshka support. 572M parameters.

Open-source embedding model with improved performance. 137M parameters.

Vision-Language-Action model for robotic control. 55B parameters.

Open-source generalist robot policy for dexterous manipulation.

Hybrid SSM-Transformer model with 256K context for enterprise RAG. 398B total / 94B active.

Compact Jamba for fast enterprise workloads. 52B total / 12B active parameters.

70B model fine-tuned with RLHF, excelling as reward model and assistant. 70B parameters.

Large-scale model optimized for enterprise synthetic data generation. 340B parameters.

Frontier multimodal LLM matching proprietary models on vision-language tasks. 72B parameters.

NVIDIA-optimized Llama 3.1 70B with enhanced helpfulness. 70B parameters.

One of the largest open-source autoregressive models. 180B parameters.

Strong open-source model that topped HuggingFace leaderboard. 40B parameters.

Compact Falcon model for efficient deployment. 11B parameters.

Base Falcon model on RefinedWeb dataset. 7B parameters.

Latest Falcon generation with improved training methodology. 10B parameters.

First large-scale pure Mamba architecture model. 7B parameters.

Open multilingual model supporting 46 languages, created by 1000+ researchers. 176B parameters.

Instruction-tuned BLOOM for cross-lingual zero-shot task generalization. 176B parameters.

Open-source code LLM on The Stack v2, supporting 600+ languages. 15B parameters.

Compact code model for fast code generation. 7B parameters.

Small code model for local development tools. 3B parameters.

Original open-source code model trained on permissive data. 15.5B parameters.

Tiny but capable model for on-device AI. 1.7B parameters.

Ultra-compact model for embedded applications. 360M parameters.

Smallest capable instruction-following model. 135M parameters.

Compact vision-language model for image and video understanding. 2.2B parameters.

State-of-the-art open image generation with exceptional prompt following. 12B parameters.

Fastest FLUX variant for real-time image generation. 12B parameters.

Premium FLUX model with highest quality outputs. 12B parameters.

Open-source TTS generating realistic speech, music, and sound effects.

Controllable text-to-speech with natural speaker descriptions. 2.3B parameters.

TTS model generating realistic dialogue with emotions. 1.6B parameters.

Real-time speech-to-speech foundation model for natural conversation. 7B parameters.

Novel two-stage TTS model excelling at prosody and expression.

Flow-matching based text-to-speech with zero-shot voice cloning.

Personal AI assistant designed for emotional intelligence and supportive conversations.

97.5% of GPT-4 performance with strong EQ capabilities.

Frontier multimodal model processing text, images, video, and audio natively.

Fast multimodal model for cost-effective applications. 21B parameters.

Compact multimodal model for edge deployment. 7B parameters.

Advanced reasoning model with real-time web search and citations.

Enhanced search-augmented model with grounded, cited answers.

Fast search-augmented model providing answers with web citations.

Bilingual model with strong tool-use and 128K context. 9B parameters.

Open-source text-to-video model with strong generation quality. 5B parameters.

Text-to-image model with relay-based generation for high quality.

Compact Apple language model for on-device AI. 1.1B parameters.

Autoregressive vision encoder pre-trained with multimodal objectives. 304M parameters.

Large action model for function calling and multi-step agentic tasks. 70B parameters.

Compact action model excelling at tool use and function calling. 8B parameters.

Multi-turn program synthesis model for code generation. 7B parameters.

Tencent's largest open-source MoE model. 389B total / 52B active parameters.

Open-source video generation model with strong physical understanding. 13B parameters.

3D asset generation model creating high-quality meshes from images/text.

Baidu's flagship model with strong Chinese-English bilingual capabilities.

Text-to-image model from Baidu with knowledge-enhanced generation.

Open-source model trained from scratch with commercial license. 30B parameters.

Commercially usable open-source base model. 7B parameters.

Research model suite for studying LLM training dynamics. 12B parameters.

Large open-source autoregressive model. 20B parameters.

Early open-source GPT alternative on The Pile. 2.7B parameters.

Powerful model with 4M token context window. 456B total parameters.

Lightning-fast text model with hybrid attention architecture.

Bilingual model excelling at instruction following. 32B parameters.

Compact bilingual model for diverse applications. 7.8B parameters.

Small but capable bilingual model. 2.4B parameters.

Fine-tuned LLaMA model trained on ShareGPT conversations. 33B parameters.

Popular fine-tuned chat model based on LLaMA. 13B parameters.

Compact chat model fine-tuned on user conversations. 7B parameters.

DPO-aligned model based on Mistral 7B with strong chat abilities. 7B parameters.

Instruction-tuned Mixtral with strong general performance. 46.7B total parameters.

Strong instruct model based on Yi-34B. 34B parameters.

High-quality open model rivaling ChatGPT. 7B parameters.

Intel-optimized chat model based on Mistral 7B. 7B parameters.

Depth upscaled model with strong performance. 10.7B parameters.

Strong open-source model with excellent tool use. 20B parameters.

Compact model with advanced reasoning and tool use. 7B parameters.

Open-source multimodal model matching GPT-4o on vision tasks. 78B parameters.

Strong bilingual model with emphasis on Chinese NLP. 13B parameters.

Compact bilingual model for Chinese-English tasks. 7B parameters.

Bilingual chat model with function calling support. 6B parameters.

Bilingual model from Beijing Academy of AI. 70B parameters.

Fully open-source bilingual model with transparent training. 7B parameters.

Fully open model with open data, weights, and code by Allen AI. 13B parameters.

Compact fully open model for research and deployment. 7B parameters.

State-of-the-art post-trained open model by Allen AI. 70B parameters.

Open multimodal model rivaling proprietary models on vision tasks. 72B parameters.

Open-source robotics framework with pretrained manipulation policies.

Medical-specialized model with expert-level clinical reasoning.

Biomedical domain model fine-tuned on PubMed data. 7B parameters.

Mistral's specialized coding model for code generation, review, and debugging.

Medical LLM adapted from Llama 2 for healthcare. 70B parameters.