AI Model Leaderboard

Advanced reasoning model with chain-of-thought capabilities and extended thinking time

Large-scale mixture of experts model with 355B total parameters and 32B active

Large-scale MoE model with 235B total parameters and 22B active, optimized for instruction following

Lightweight version of GLM 4.5 with 106B total parameters and 12B active

Second generation model with improved reasoning and extended context

Large-scale model optimized for synthetic data generation and instruction following

Enhanced bilingual model with improved reasoning and longer context

Early large language model from Nvidia focused on enterprise applications

Advanced code generation model with 4x larger context than StarCoder

First generation model from StepFun with focus on Chinese language understanding

Test model for benchmarking purposes with exceptional performance across all metrics

Bilingual Korean-English model with strong performance on both languages

Revolutionary reasoning model achieving near-human performance on ARC-AGI

Flagship multimodal model from Mistral AI with 405B parameters, designed for advanced reasoning and instruction following

Massive-scale coding-specialized model with 480B parameters, designed for complex software development tasks

Advanced Llama 3.3-based model enhanced by Nvidia with superior reasoning and instruction following capabilities

Open-source reasoning-focused model built on Nemotron architecture with enhanced logical thinking capabilities

Multimodal voice and text model from Mistral AI with advanced audio processing and speech understanding capabilities

Instruction-tuned version of DBRX with enhanced capabilities

Advanced reasoning-focused MoE model with enhanced thinking capabilities and step-by-step problem solving

Specialized coding model from Mistral AI with 22B parameters, optimized for software development and programming tasks

Advanced Chinese-English bilingual model with exceptional long-context understanding and instruction following

SK Telecom's vision-language model optimized for Korean and English

Efficient coding-specialized model with 30B parameters, optimized for software development and programming tasks

132B parameter MoE model outperforming GPT-3.5 and competitive with GPT-4

Kakao's Korean-optimized language model with strong instruction following

Base version of Kimi K2 model optimized for research and fine-tuning applications with strong foundational capabilities

Largest open-source model rivaling proprietary systems

Advanced MoE model with 30B total parameters and 3B active, optimized for instruction following with high efficiency

Open language model optimized for OCR and document understanding

Compact reasoning-focused model with 7B parameters, optimized for efficient deployment with strong logical capabilities

Flagship model with exceptional coding capabilities

Llama 3.3 70B delivers performance nearly matching the 405B model while being significantly more efficient and cost-effective.

180B parameter model, top open-source model at launch

Leading Chinese model with strong math and coding

Large MoE model with efficient architecture

Advanced multimodal model with strong reasoning capabilities

GPT-4o mini is OpenAI's most cost-efficient small model, designed for fast responses and handling large contexts.

Stability AI's 12B parameter language model with improved performance

Arcee AI's efficient 4.5B parameter model focused on domain adaptation

Highly capable open model with extended context

Claude 3.5 Haiku is Anthropic's fastest and most cost-effective model, optimized for speed while maintaining strong performance.

Large mixture-of-experts model with strong performance

Efficient transformer alternative with state-space model architecture

Balanced model striking ideal balance between intelligence and speed for enterprise workloads

First multimodal Llama model with vision capabilities

Advanced multimodal model with unprecedented 2M token context and superior reasoning

Bilingual model excelling in English and Chinese

Most capable Grok model, available at premium tier for $300/month

Conversational AI with emotional intelligence

Advanced iteration of DeepSeek-R1 with enhanced reasoning and larger scale

Enhanced Kimi model with dramatically expanded context window and improved reasoning across all domains

Cost-efficient reasoning model with performance approaching larger models

Efficient reasoning model with o3-level capabilities at lower cost

Fastest Claude model optimized for speed and efficiency

Reasoning-focused model with chain-of-thought capabilities

Enterprise model optimized for RAG and tool use

DeepSeek-R1 is an advanced reasoning model that rivals OpenAI o1 performance through reinforcement learning and chain-of-thought reasoning.

Fine-tuned Mistral-7B with improved helpfulness and harmlessness

OpenAI o1 is a reasoning model that uses chain-of-thought processing to solve complex problems with unprecedented accuracy.

Next-generation Qwen model with significantly expanded scale and capabilities

Early release version of Mixtral 8x22B with mixture-of-experts architecture, showing the initial capabilities before optimization

Most capable Claude model with enhanced reasoning and extended context

Balanced Claude 4 model offering Opus-level capabilities with better efficiency

Next-generation model from xAI with enhanced reasoning and real-time capabilities

Deep-thinking reasoning model with multimodal capabilities, designed for complex problem solving

Advanced model with 10x more compute than Grok 2, featuring reflection capabilities

Enhanced Sonnet model bridging the gap to Claude 4

Gemini 2.0 Pro is Google's most capable multimodal AI model, designed for the agentic era with advanced reasoning and tool use capabilities.

Lightning-fast multimodal model optimized for speed and efficiency

Most powerful GLM model with breakthrough 1M token context window and enhanced capabilities

Large Mixture of Experts model trained on 20T+ tokens with extended context capabilities

o1-mini is a cost-efficient reasoning model optimized for STEM tasks, particularly mathematics and coding, with performance matching o1 on technical benchmarks.

Third generation A.X model with significantly improved reasoning capabilities and expanded context window for enterprise applications

14B parameter model specialized in math and complex reasoning

Pioneering open-source mixture-of-experts model

Small language model with outsized performance

Native multimodal model with Mixture of Experts architecture, supporting text, image, audio, and video

Advanced model with real-time information access

Lightweight multimodal model for on-device deployment

Most capable Claude model with superior coding and reasoning

Specialized model for ultra-long context processing with 2M token window

40B parameter model optimized for performance and efficiency

OpenAI's flagship omni-model with native multimodal capabilities

First generation Kimi model with exceptional long context capabilities and strong Chinese-English bilingual performance