Wafer says it beat Cerebras on latency for YC's AI Office Hours

Wafer says it served Y Combinator's AI Office Hours, where users get startup advice from AI versions of YC partners, faster than Cerebras did. By Wafer's account, GLM-5.2 on its endpoint averaged 379 milliseconds of latency against 674 milliseconds for Gemma 4 31B on Cerebras, a 44% cut with a much larger model. YC had tested lightweight Gemma and OpenAI models before moving to a dedicated Wafer endpoint. Wafer says its own agents tuned the serving setup to YC's request rate, cache usage and prompt and response lengths, and that users spent 2.5 more minutes talking on Wafer than with other providers. These are vendor figures from a customer story, not an independent benchmark. The useful point for builders is that serving tuned to a specific workload can outweigh raw model size when conversational latency drives engagement.