Submissions by Gcam

From GPT-4 to Mistral 7B, there is a 300x range in the cost of LLM inference

Show HN: LLM Benchmarks Leaderboard with 60 model and API host combinations

Mistral API reduces time to first token by 10x (only place for Mistral Medium)

240 Tokens/s achieved by Groq's custom chips on Lama 2 Chat (70B)

New GPT-4 Turbo (0125 Preview) slightly faster per initial benchmarks