### Answer I don't have access to community discussions or user opinions about which LLMs perform best within Retell AI. The sources I have don't contain community feedback or user testimonials comparing model performance. However, here's what's available in Retell AI for LLM selection: 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

[(1) ]() Speech-to-Speech Models: -

gpt-4o-realtime

,

gpt-4o-mini-realtime

-

gpt-realtime

,

gpt-realtime-mini

General Guidance from Retell AI: - For function calling and data collection, use lower temperature (0.0–0.3) for more deterministic responses [(2) ]() - GPT-4o achieved 88.7 on the MMLU benchmark and sets records in audio speech recognition [(3) ]() - Claude excels in understanding nuanced instructions and prioritizes safety For real community feedback, I'd suggest asking directly in this channel or checking past discussions here.

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] Get Retell LLM](https://docs.retellai.com/api-references/get-retell-llm) [[2] Configure LLM Options](https://docs.retellai.com/build/llm-options) [[3] Choosing the Best LLM for Your AI Voice Agents | Retell AI](https://www.retellai.com/blog/choosing-the-best-llm-for-your-voice-ai-agents) ​

### Answer The sources I have don't contain explicit documentation explaining the difference between "Text LLM Models" and "Speech-to-Speech Models" within Retell AI's configuration options. What I can share from general Retell AI architecture: The standard voice AI pipeline involves three separate components: - Speech-to-Text (STT) : Captures audio and converts it to text [(1) ]() - LLM Processing: The text is fed into a language model to generate a response - Text-to-Speech (TTS) : Converts the response text back into spoken audio Text LLM Models likely follow this traditional pipeline—audio is transcribed to text, processed by the LLM, then synthesized back to speech. Speech-to-Speech Models (like

gpt-4o-realtime

) are designed to process audio input and generate audio output more directly, enabling lower latency and more natural real-time interactions. [(2) ]() For specific technical details on how these differ in Retell AI's implementation, I'd recommend checking the [LLM Options documentation]() or asking here for community insights.

​ To learn more, see the sources I considered: [[1] Vapi AI Review 2025: Pricing, Features & Top Alternative | Retell AI](https://www.retellai.com/blog/vapi-ai-review) [[2] Choosing the Best LLM for Your AI Voice Agents | Retell AI](https://www.retellai.com/blog/choosing-the-best-llm-for-your-voice-ai-agents) ​

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

### Answer Based on the sources I have, here's what community members have shared about Gemini and Claude options: Gemini 2.5 Flash: - One user reported issues when switching from GPT-4.1 to Gemini 2.5 Flash, experiencing 500 errors with "Error in LLM call" during tool call transitions in their flow. [(1) ]() - A community member noted that "Gemini 2.5 have some issues with tool calls that may effect the agent" and suggested trying GPT or Claude instead. Claude 4.5 Sonnet: - Anthropic has "significantly improved subject matter expertise in finance, healthcare, law, and STEM" with this model. [(2) ]() - It's recommended for voice agents handling "specialized conversations like medical scheduling, financial advisory, legal intake." - Claude excels in understanding nuanced and complex instructions, including humor and subtle context. [(3) ]() - It prioritizes safety and ethical alignment, making it suitable for sensitive applications. General takeaway: Based on these discussions, Claude appears to have more positive feedback for specialized/compliant use cases, while Gemini 2.5 Flash has had some reported tool calling issues. If you're experiencing problems with Gemini, the community suggestion was to try GPT or Claude alternatives.

​ To learn more, see the sources I considered: [[1] 2.5 Flash Fails to Extract a Variable](https://discord.com/channels/1186424376300679209/1399142019145924709) [[2] Claude Sonnet 4.5 is now live on Retell AI](https://retell.discourse.group/t/claude-sonnet-4-5-is-now-live-on-retell-ai/118) [[3] Choosing the Best LLM for Your AI Voice Agents | Retell AI](https://www.retellai.com/blog/choosing-the-best-llm-for-your-voice-ai-agents) ​

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

### Answer ### Answer I don't have access to community discussions or user feedback specifically about how emotional people are making their AI voices for sales or customer service calls. What the documentation covers about voice emotional capabilities: - Voice cloning allows agents to use human-sounding voices reflecting your desired tone, pitch, cadence, accent, and style. [(1) ]() - Text-to-speech advancements enable AI voice agents to "replicate tone, emotion, and speech patterns for a more engaging and authentic customer experience." - This "boosts trust and customer satisfaction, making AI voice agents nearly indistinguishable from human agents in conversation." Emotional intelligence at scale: - AI voice agents can "express patience, warmth, and tone in ways that sometimes outperform humans." [(2) ]() - This is especially important for "sensitive calls like collections or healthcare reminders," "high-stress customer support scenarios," and "brand-critical sales and onboarding interactions." - Because they "never tire, forget, or lose their temper," AI voice agents can deliver "consistent, emotionally intelligent experiences at scale." For specific community insights on voice emotion settings people are using, I'd suggest asking directly in this channel.

​ To learn more, see the sources I considered: [[1] AI Voice Agents in 2025: Everything Businesses Need to Know | Retell AI](https://www.retellai.com/blog/ai-voice-agents-in-2025) [[2] 2025 B2B Guide to AI Phone Calls & Voice Agents | Retell AI | Retell AI](https://www.retellai.com/blog/b2b-guide-to-ai-phone-calls) ​

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

### Answer ### Answer Voice Temperature controls how stable or variant the voice is, with values ranging from [0, 2]. [(1) ]() - Lower values = more stable voice output - Higher values = more variant speech generation This setting currently only applies to ElevenLabs (

11labs

) voices. If unset, the default value of 1 will apply. What community members have shared: One user asked about the relationship between Retell's voice temperature and ElevenLabs settings. The Retell team confirmed it's "not exactly equal" but "the setting is based on ElevenLabs's stability setting." [(2) ]() For making voices sound less robotic, a community member recommended experimenting with: [(3) ]() - **Voice temperature**: Set higher for more natural variation - **Voice speed**: Adjust for faster or slower speech - **Voice volume**: Tune to optimal loudness - **Voice model**: Try different models your voice provider supports I don't have specific community data on what voice temperature values people are using for sales or customer service calls. For those insights, I'd suggest asking directly in this channel.

​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] Voice Temperature for ElevenLabs Voices](https://discord.com/channels/1186424376300679209/1237472202002268210) [[3] Voice & Speech Settings](https://discord.com/channels/1186424376300679209/1341458231196782613) ​

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