LLM Latency Leaderboard

Methodology

• Measures time-to-first-token for serverless LLM providers
• Measured from EU Central, observed similar results with US Central
• _{Pass rate: A simple URL classification task was used with a small max_tokens limit across multiple samples. (Some reasoning models failed by initially outputting <think> tokens.)}
• Discussion on Reddit.