​ To learn more, see the sources I considered: [[1] List Retell LLMs](https://docs.retellai.com/api-references/list-retell-llms) [[2] Get Retell LLM](https://docs.retellai.com/api-references/get-retell-llm) [[3] Create Retell LLM](https://docs.retellai.com/api-references/create-retell-llm) [[4] Platform Changelogs | Retell AI](https://www.retellai.com/changelog) [[5] Performance-related configurations and noticeable latency during live calls](https://community.retellai.com/t/performance-related-configurations-and-noticeable-latency-during-live-calls/747) ​

### Answer Based on the available models, here's how they compare for performance: [(1) ]() [(2) ]() [(3) ]() Intelligence ranking: GPT-5.2 > GPT-5.1 > GPT-5 > Claude-4.5-Sonnet > GPT-4.1 > GPT-5-mini > Claude-4.5-Haiku > Gemini-3.0-Flash > GPT-4.1-mini > Gemini-2.5-Flash > GPT-5-nano > Gemini-2.5-Flash-Lite > GPT-4.1-nano [(4) ]() Latency (lower is faster) : GPT-4.1-nano < GPT-5-nano < Gemini-2.5-Flash-Lite < GPT-4.1-mini < Gemini-2.5-Flash < GPT-5-mini < Gemini-3.0-Flash < Claude-4.5-Haiku < GPT-4.1 < GPT-5 < Claude-4.5-Sonnet < GPT-5.1 < GPT-5.2 For best performance balance, GPT-4.1 is recommended as it offers strong intelligence with sub-800ms latency. [(5) ]() If you need even faster responses, GPT-4.1-mini or GPT-4.1-nano provide lower latency while maintaining good performance. [(4) ]() You can also enable the Fast Tier option (formerly "High Priority") for 1.5x model cost to get predictable latency, uncapped scale, and higher reliability. For speech-to-speech models, you can use

gpt-4o-realtime

,

gpt-4o-mini-realtime

,

gpt-realtime

, or

gpt-realtime-mini

. [(1) ]() [(2) ]() [(3) ]()

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

### Answer Based on user experience from the forums, here's what community members have shared about model performance: For latency issues, users experiencing 3000ms+ latency found that switching to Claude 4.5 Haiku provided better performance than Gemini models, though it still showed 2403ms in one case. [(1) ]() Users also reported that GPT 4.1 Fast Tier delivered 2652ms, which wasn't optimal but was an improvement. GPT 3.5 Turbo was mentioned as "supposed to be fastest" for LLM latency, [(2) ]() though one user experienced median latency of 3200ms when using the Assistants API, which was noted to be "quite slow when integrated with voice layer." GPT 4.1 was recommended by the community for better accuracy, with suggestions to test it when dealing with knowledge base issues and hallucinations. [(3) ]() The model has a 1M token context window compared to GPT 4o's 128K. Key recommendations from forum users: - Change to a faster LLM to reduce response time and improve performance [(4) ]() - Use the fast tier or switch LLM providers if LLM latency is consistently above 900ms [(4) ]() [(1) ]() - Avoid using the Assistants API for voice applications due to structural slowness [(2) ]() The forums show users actively testing multiple models

, with GPT 4.1 and Claude 4.5 Haiku being the most discussed for balancing performance and accuracy based on real-world usage.

​ To learn more, see the sources I considered: [[1] Slow Latency (3000ms, triple what it should be)](https://community.retellai.com/t/slow-latency-3000ms-triple-what-it-should-be/591) [[2] llm latency](https://discord.com/channels/1186424376300679209/1247142177574879264) [[3] Discord message from @leo_poeppel_74864 on 2025-05-07](https://discord.com/channels/1186424376300679209/1186462578310910112/1369575933668888590) [[4] Performance-related configurations and noticeable latency during live calls](https://community.retellai.com/t/performance-related-configurations-and-noticeable-latency-during-live-calls/747) ​

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

ncy scores for each model, I don't have that information in the available sources. Would you like me to provide a general answer based on broader knowledge?

### Answer Based on the available sources, I can provide the model options that Retell AI supports, but specific intelligence and latency scores for each model are not documented in the sources provided. [(1) ]() [(2) ]() [(3) ]() Available text LLM models: - gpt-4.1 - gpt-4.1-mini - gpt-4.1-nano - gpt-5 - gpt-5.1 - gpt-5.2 - gpt-5-mini - gpt-5-nano - claude-4.5-sonnet - claude-4.5-haiku - gemini-2.5-flash - gemini-2.5-flash-lite - gemini-3.0-flash Available speech-to-speech models: - gpt-4o-realtime - gpt-4o-mini-realtime - gpt-realtime - gpt-realtime-mini Performance information from sources: Retell AI typically achieves sub-600ms to 620ms average end-to-end latency with optimized configurations. [(4) ]() [(5) ]() Specific performance metrics documented: [(5) ]() - Time-to-First-Token: 180ms average - End-to-End Latency: 620ms average - Barge-in Response: 140ms average The sources indicate that if you need to reduce latency, you can: - Change to a faster LLM to reduce response time - Change to a faster TTS for quicker audio output - Enable the model_high_priority setting (high priority pool) for lower and more consistent latency [(1) ]() [(2) ]() [(3) ]() For detailed intelligence rankings and specific late

​ To learn more, see the sources I considered: [[1] List Retell LLMs](https://docs.retellai.com/api-references/list-retell-llms) [[2] Create Retell LLM](https://docs.retellai.com/api-references/create-retell-llm) [[3] Get Retell LLM](https://docs.retellai.com/api-references/get-retell-llm) [[4] ](https://www.retellai.com/resources/retell-ai-receptionist-understands-calls-better) [[5] Performance-related configurations and noticeable latency during live calls](https://community.retellai.com/t/performance-related-configurations-and-noticeable-latency-during-live-calls/747) ​

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