Model Roles and LLM Access
Whisperer uses several specialized models through a unified AI gateway — one per task. This approach lets the system select the optimal model for each type of request: a fast chat response, code generation, screenshot analysis, System Design, or embeddings.
When to Read This
Read this article to understand:
- which models are working under the hood;
- how plans differ in terms of model access;
- how to choose the right model for your needs.
Seven Model Roles
| Role | Purpose | Models (examples) |
|---|---|---|
| responses | Chat answers to questions during a meeting | Claude Sonnet/Haiku, GPT-4o mini, DeepSeek |
| coding | Code explanation and snippet generation | DeepSeek Coder, Claude Sonnet, GPT-4o |
| vision | Screenshot and image analysis | Claude Sonnet with vision, GPT-4o Vision |
| system_design | Structured response with Mermaid diagrams | Claude Opus, GPT-4o |
| generation | Mind-map enrichment, node generation | Claude Haiku, GPT-4o mini |
| transcription | Speech recognition (Whisper) | Whisper |
| embedding | Vectorization of knowledge base notes for RAG | Embedding model |
Access by Plan
| Plan | Available Roles | Limitations |
|---|---|---|
| Free | responses (basic models) | Basic tier only; 60 min/month |
| Start / Pro | responses, coding, vision, system_design, generation | Wide model selection; no frontier models |
| Max | All roles + frontier models | Claude Opus, flagship GPT, top DeepSeek |
Frontier models are the most powerful and expensive models from each provider (Claude Opus, GPT-4.1, DeepSeek R1, and equivalents). Available exclusively on the Max plan.
How to Change the Model
- Open Settings in the client or web dashboard (
/app/settings). - Go to the Models section.
- For each role, select a model from the set available on your plan.
- Save — the settings apply to the next session.
📸 [Screenshot: settings screen with model dropdowns per role]
Available Model Families
Models from the following providers are available:
- Anthropic: Claude Opus, Sonnet, Haiku
- OpenAI: GPT-4o, GPT-4o mini, GPT-4.1
- DeepSeek: DeepSeek V3, Coder, R1
- Alibaba: Qwen
- Google: Gemini Flash, Pro
- xAI: Grok
- Meta: Llama 3
The available model set is updated as new models are added.
Common Errors
| Error | Cause | Fix |
|---|---|---|
| Model does not respond | The selected model is temporarily unavailable | Switch to another model in the same role |
| "Frontier model unavailable" | Your plan does not include frontier models | Upgrade to Max |
| Response is unexpectedly short | A model with a small context window was selected | Switch to a more capable model |
Best Practices
- For quick hints during an interview use Haiku / GPT-4o mini — responses arrive faster.
- For System Design choose Claude Opus or GPT-4o — they handle structured diagrams better.
- For code DeepSeek Coder is often more accurate than other models of comparable size.
- Max plan with frontier models is especially valuable for non-standard situations that require maximum reasoning.