Neutral, receipted benchmarks, re-provable on demand.
RedCrown runs the same pinned tasks through each model in the set under identical conditions, scores them on a real labeled set, and names the cheapest model that clears the quality bar. Every number carries a receipt, and the answer is re-checked as prices and models change. No vendor's thumb on the scale.
Frontier vs open-source LLMs
GLM, Claude, GPT, Grok, Gemini, DeepSeek, Qwen, Llama and Kimi head to head on a pinned MMLU-Pro subset, ranked by accuracy and dollars per correct answer.
View the benchmark →Clinical transcription
Speech-to-text models on a public clinical corpus, scored on word error rate, medical-term F1, and cost per minute, with per-clip receipts.
View the benchmark →Each benchmark runs on the published redcrown CLI, prices each model at its published list rate, counts and shows errors, and links a public proof you can check item by item for every published run. Run benchmarks anywhere; prove them here.