# Choosing a model

Canonical page: https://www.questionpunk.com/support/ai-choosing-a-model

How to pick the right AI model for your study.

Time: 5-15 minutes

Level: Intermediate

Audience: Researchers, Product teams

## Steps

### 1. Understand the options

QuestionPunk supports 150+ models across 20+ providers. New text AI interviews default to **OpenAI GPT-5.4 Mini**, while new Pipeline voice interviews default to **Claude Sonnet 5**. Claude Haiku 4.5 remains the generic default, and existing surveys keep their saved model.
**Anthropic Claude:** Claude Haiku 4.5 (fast and economical), Sonnet 4.6 and Sonnet 5 (balanced speed and depth), Opus 4.6 through Opus 5 (richest follow-ups for deep qualitative work), and Fable 5. Thinking-enabled variants are also available for extended reasoning.
**OpenAI:** GPT-5.6 Luna/Sol/Terra (latest generation), GPT-5.5/Pro, GPT-5.4/Pro/Mini/Nano, plus GPT-5.3 Codex, 5.2, 5.1, and 5 series models, GPT-4.1 family, GPT-4o, o3, o3 Mini, o3 Pro, o4 Mini, gpt-oss open-weight models, and specialized Codex and Chat variants.
**Google:** Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash/Lite, Gemini 3.1 Pro/Flash Preview, Gemini 2.5 Flash/Pro, plus Gemma 4 and Gemma 3 open models.
**Meta:** Llama 4 Maverick, Llama 4 Scout, Llama 3.3, and 3.2 models.
**Mistral:** Medium 3.5, Large 3, Medium 3.1, Small 4, Nemo, Ministral 3, Devstral, Codestral, and Saba models.
**DeepSeek:** V4 Pro/Flash, V3.2, V3.1, R1 series.
**xAI:** Grok 4.6 and Grok 4.5.
**Amazon:** Nova Pro, Lite, and Micro.
**Moonshot:** Kimi K3.
**Cohere:** Command A.
**Qwen, Nvidia, StepFun, Arcee, Upstage, Zhipu,** and others are also available, including free-tier and zero-cost options.

### 2. Select in survey settings

Choose the model in the AI interviewer settings for each interview question. You can set different models for different interview questions within the same survey.

### 3. Pilot and compare

Run a short A/B pilot with different model settings to compare output quality before sending to your full sample.

## Details

QuestionPunk supports 150+ models across 20+ providers including Anthropic, OpenAI, Google (Gemini), Meta (Llama), Mistral, DeepSeek, xAI (Grok), Amazon (Nova), Moonshot (Kimi), Cohere, Qwen, Nvidia, StepFun, Arcee, Upstage, and Zhipu. New text AI interviews default to OpenAI GPT-5.4 Mini, while new Pipeline voice interviews default to Claude Sonnet 5. Claude Haiku 4.5 remains the generic default, and existing surveys keep their saved model.

For most studies, Claude Haiku 4.5 or Sonnet 4.6 provides the best balance of quality and speed. Opus 4.6 through Opus 5 deliver the richest, most nuanced follow-ups for complex qualitative research. Thinking-enabled model variants are available for tasks that benefit from extended reasoning.

Free-plan users have access to Claude Haiku 4.5, OpenAI GPT-5.4 Mini, and Google Gemma 4 31B. Haiku and GPT-5.4 Mini consume AI credits at low rates, while Gemma 4 31B is completely free (zero credit cost). Many other open-weight models — including Llama 4, Mistral Small, and Nvidia Nemotron — are available at very low cost on paid plans. Premium models from Anthropic and OpenAI offer the highest quality.

Run a short pilot with 5-10 respondents using different model settings to compare output quality before committing to your full sample.

## Related articles

- [How AI interviews work](https://www.questionpunk.com/support/ai-how-it-works.md)
- [Tips for prompts](https://www.questionpunk.com/support/ai-prompt-tips.md)