### Answer Based on the available sources, here's what I found regarding token recommendations for voice agents: Recommended prompt size: Under 1,000 tokens One community member shared their approach: "I kind of try to keep my prompt size below 1,000 tokens... because after that it seems to hallucinate more." [(1) ]() Billing threshold: 3,500 tokens LLM usage is billed based on token count, with prompts up to 3,500 tokens following the existing flat rate. Prompts exceeding 3,500 tokens will scale proportionally in cost. [(2) ]() Important consideration: The 3,500 token limit isn't just your prompt—it also includes the transcript between agent and user, plus tool call history and results. So even if your prompt is under 3,500 tokens, the total can exceed that threshold during a conversation. [(3) ]() Summary: - Keep prompts under 1,000 tokens for better accuracy and reduced hallucination - Be aware that 3,500 tokens is the billing threshold before surcharges apply - Factor in that conversation history and tool calls add to your token count

​ To learn more, see the sources I considered: [[1] Master The Future of Voice AI - Conversation Design & Agent Persona]() [[2] Platform Changelogs | Retell AI](https://www.retellai.com/changelog) [[3] Token Surcharge when prompt tokens is under 3500](https://discord.com/channels/1186424376300679209/1442646663335510179) ​

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

### Answer With an 8,000 token prompt, here's what you should know: Billing Impact Your prompt exceeds the 3,500 token threshold, so you'll be charged at a scaled rate. [(1) ]() The calculation works like this: - Scaling Factor = 8,000 ÷ 3,500 = 2.29 - For a 60-second call, you'd be billed for approximately 138 seconds instead of 60 Stability Concerns According to feedback from Retell engineers: "Using over 33,000 tokens can significantly reduce stability, causing the LLM to become confused and respond inaccurately. Generally, the prompt size should remain below 8,000 tokens to ensure optimal performance. Larger prompts also increase latency and costs." [(2) ]() Token Limit The LLM prompt token limit is set to 32,768 tokens, so you're within the hard limit. [(3) ]() Recommendations At 8,000 tokens you're right at the threshold engineers suggest for optimal performance. Consider: - Trimming your prompt if possible to reduce costs and improve stability - Moving detailed information to a Knowledge Base instead of the prompt - Being aware that conversation history and tool calls will add to your token count during calls [(1) ]()

​ To learn more, see the sources I considered: [[1] Exceptions to Our Per-Minute Pricing](https://docs.retellai.com/accounts/billing-exceptions) [[2] Prompt exceeds the limit of 32768 tokens, please contact support if you need longer context.](https://discord.com/channels/1186424376300679209/1432544872421789789) [[3] Platform Changelogs | Retell AI](https://www.retellai.com/changelog) ​

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