Seeking LLM alternative to Claude Haiku for psychology ed AI

MaD1254 · reddit · 2026-08-22

A developer is building a voice-based AI project for psychology education where students interview an AI-simulated patient. The current stack uses Sarvam (STT/TTS) and Claude Haiku 4.5 (LLM). While the MVP works, the author is concerned about scaling costs. For TTS, the author prefers ElevenLabs over Sarvam for more natural English speech (pauses, hesitation, emotion). For the LLM, the author needs a model capable of following complex patient profiles, maintaining personality/emotion, handling long conversations, and providing psychologically believable responses. The author is considering Gemini, GPT, Kimi, MiniMax, and open-source models and seeks recommendations.

Original post →

More from Apps

Apps channel →