Hey everyone I’ve been testing different providers in Vapi and noticed something that doesn’t quite match the displayed metrics. When using: Provider: OpenAI Model: GPT 5.1 Instant I get around: Cost: ~$0.18/min Latency: ~975 ms And the assistant actually feels fast and responsive. However, when switching to: Provider: Anthropic (Bedrock) Model: Claude Sonnet 4 It shows: Cost: ~$0.10/min Latency: ~275 ms But in practice, the assistant takes noticeably longer to respond — significantly slower than GPT 5.1 Instant. Is this expected? Are those latency numbers measured differently (e.g. first token vs full response), or could there be something else affecting real-world performance? Would love some clarification here because the perceived latency doesn’t match the reported metrics at all. Thanks! 🚀