Model Benchmarks
Real outputs instead of synthetic eval scores. Each benchmark gives every model the same prompts through the same pipeline, then shows you everything that came out, including the failures.
LLM Tier List Maker
Rank this week's most-used paid models on OpenRouter alongside the Artificial Analysis leaders. Add new releases the day they drop and export a shareable image.
Make your tier listQwen Image 2.1: INT4 + FP4
I tested INT4 on a 4070 Ti SUPER and FP4 on a 5090. Compare images, model sizes and measured speed against BF16.
Compare the imagesAI Video Generation
24 LLMs each generate 7 Remotion motion-design videos from identical prompts via FrameCall. The lineup includes GPT-5.6, Claude, Gemini 3.1 Pro, DeepSeek V4 and Kimi K2.6.
Watch the resultsOpenRouter Model Spend
Weekly USD spend per model on OpenRouter over the last 52 weeks, priced with time-accurate rates plus cache-adjusted estimates. Refreshes weekly.
See who pays the mostClaude Code Performance
Daily SWE-Bench-Pro pass rates for Claude Code on Opus 4.6, measured against a fixed baseline so 1-day, 7-day and 30-day regressions show up as they happen. Data from Marginlab, refreshed every 6 hours.
Check today's pass rate