Artificial Analysis

Artificial Analysis is a technical evaluation platform that systematically benchmarks and compares large language models (LLMs) and their API providers. The site maintains a comprehensive leaderboard of over 100 AI models sourced from major developers such as OpenAI, Google, DeepSeek, Anthropic, Meta, and others. Each model is assessed across multiple quantitative dimensions, including intelligence scores (based on standardized test suites), output speed measured in tokens per second, latency (time to first token, or TTFT), pricing per token, and context window size. The platform also aggregates these metrics into a unified ranking that allows users to quickly identify the most capable, fastest, or most cost-effective model for a given use case. Beyond the core LLM leaderboard, the site extends its analysis to specialized domains: it evaluates models for coding tasks, speech recognition and synthesis, image generation, video processing, music generation, and hardware performance. A dedicated "API Providers Leaderboard" compares the performance and reliability of cloud-hosted inference services, helping developers choose between providers based on real-world latency, throughput, and cost stability. The platform includes interactive features such as "Arenas," where users can simulate head-to-head comparisons between arbitrary models across selected metrics. Each model entry is accompanied by detailed methodology notes, including the evaluation benchmarks used (e.g., MMLU, HumanEval, GSM8K) and the exact test conditions (temperature, batch size, hardware). Users can filter and sort the leaderboard by any metric, enabling fine-grained analysis—for example, finding the cheapest model with a context window above 128K tokens and output speed exceeding 100 tokens per second. Th