### Answer To reduce your 1400-1800ms latency, work through these steps in order: [(1) ]() 1. Choose a faster LLM model Model choice is often one of the largest factors in LLM response time. Smaller models respond faster, while larger reasoning-oriented models are slower to first token. If your use case can tolerate slightly less reasoning, switching to a smaller model noticeably reduces latency. 2. Shorten your prompt Every token in your prompt adds to the time-to-first-token, and that cost is paid on every turn of the conversation. Keep system prompts tight and focused on the current task. Move rarely-needed detail into the knowledge base or into tool descriptions rather than the main prompt. Prompts beyond roughly 8k tokens become noticeably slower. 3. Enable fast tier Fast tier routes your LLM calls through dedicated, high-priority infrastructure, reducing both average response time and call-to-call variance. Additional checks: - If you're using the accuracy-optimized transcription mode, switch to the speed-optimized mode for about 200ms improvement. - Turn off Boosted Keywords if enabled, as it can add transcription latency (approximately 1s more) . [(2) ]() - Disable any turtle-icon features you don't need, and aim to keep estimated latency under 1.5s. [(1) ]() - Turn off unnecessary features like emotion in voice settings if you don't need them. Use the per-call latency breakdown in your dashboard to identify which component (LLM, ASR, or TTS) is dominating your latency, then target that specific component.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

​ To learn more, see the sources I considered: [[1] Troubleshoot high latency](https://docs.retellai.com/reliability/troubleshoot-latency) [[2] >4000 ms latency](https://discord.com/channels/1186424376300679209/1291941794640167026) ​

### Answer Here's what you should aim for based on Retell AI's performance targets: Target latency by component: [(1) ]() [(2) ]() [(3) ]() - End-to-end (e2e) : Sub-800ms is typical for Retell AI, with production agents aiming for 800ms or lower [(1) ]() [(3) ]() - LLM latency: This varies significantly by model choice, but faster models should be well under your current total [(1) ]() - ASR (transcription) : Speed-optimized mode adds about 200ms less than accuracy-optimized mode - TTS (text-to-speech) : Varies by voice provider, but should be a smaller component of total latency Performance benchmarks: [(2) ]() [(3) ]() Retell AI achieves sub-600ms to 620ms average end-to-end latency in production environments , with responses expected within 500ms for natural conversation flow [(3) ]() . How to check your breakdown: You can view detailed per-call latency breakdowns in the "Call History" section of your dashboard, which shows P50, P90, and P99 metrics for each component (e2e, asr, llm, tts) [(4) ]() . You can also retrieve this programmatically using the Get Call API . Since your current 1400-1800ms is significantly above the sub-800ms target, use your dashboard's latency

breakdown to identify which specific component (LLM, ASR, or TTS) is dominating, then apply the troubleshooting steps from my previous response to that component [(1) ]() .

​ To learn more, see the sources I considered: [[1] Troubleshoot high latency](https://docs.retellai.com/reliability/troubleshoot-latency) [[2] Bland AI Review: Real Costs, Latency Issues & Better Option | Retell AI](https://www.retellai.com/blog/bland-ai-reviews) [[3] ](https://www.retellai.com/resources/2025-best-voice-ai-companies-call-center-automation) [[4] Check actual latency](https://docs.retellai.com/reliability/check-actual-latency) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Your **transcription latency at 600-650ms is your main bottleneck**—it's consuming nearly half of your total latency. [(1) ]() Immediate fixes for transcription: 1. Switch to speed-optimized transcription mode if you're currently using accuracy-optimized mode. This saves about 200ms. 2. Turn off Boosted Keywords if enabled, as it can add transcription latency. Your other components look reasonable: - TTS (300ms) : Within acceptable range - LLM (350-500ms) : Reasonable for gpt-4.1-mini - Knowledge Base (75-125ms) : Good [(2) ]() - Response Eagerness (100ms) : Standard [(1) ]() Additional optimization: If you've already optimized transcription and still need lower latency, consider enabling fast tier, which routes your LLM calls through dedicated, high-priority infrastructure to reduce both average response time and variance. After making transcription changes, your total latency should drop closer to the sub-800ms target (300ms TTS + 350-500ms LLM + 400-450ms transcription + 75-125ms KB + 100ms eagerness = ~1225-1475ms initially, then ~1025-1275ms after transcription optimization) .

​ To learn more, see the sources I considered: [[1] Troubleshoot high latency](https://docs.retellai.com/reliability/troubleshoot-latency) [[2] Check actual latency](https://docs.retellai.com/reliability/check-actual-latency) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer After optimizing transcription, here are additional ways to reduce latency further: [(1) ]() LLM optimization: - Switch to an even faster model if your use case allows. Smaller models respond faster than larger reasoning-oriented models. Since you're using gpt-4.1-mini (350-500ms) , you could test an even lighter model to see if it meets your needs while reducing that 350-500ms window. - Enable fast tier to route your LLM calls through dedicated, high-priority infrastructure, reducing both average response time and call-to-call variance. Prompt optimization: - Shorten your prompt further. Every token in your prompt adds to time-to-first-token, and that cost is paid on every turn. Keep system prompts tight and focused on the current task. Move rarely-needed detail into the knowledge base or tool descriptions. Prompts beyond roughly 8k tokens become noticeably slower. TTS optimization: - Try a different voice provider if TTS latency matters. Latency varies by provider, so testing alternatives to Cartesia could help reduce your 300ms TTS time. - Disable emotion or other voice settings if enabled, as they can increase TTS latency in certain cases. Other optimizations: - Turn off unnecessary turtle-icon features and aim to keep estimated latency under 1.5s. - Consider geographic distance. If you're calling across countries or continents, use a phone number in the same region as your users. After applying transcription fixes and these additional optimizations, you should be able to approach the sub-800ms target that Retell AI typically achieves.

​ To learn more, see the sources I considered: [[1] Troubleshoot high latency](https://docs.retellai.com/reliability/troubleshoot-latency) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.