Answer website visitors with DeepSeek
DeepSeek V4.1 Flash is an open-weight model that reads your content thoroughly and finds the right excerpt as reliably as the best models on the list.
Every paid plan can choose DeepSeek V4.1 Flash. Free answers with Claude Sonnet 5.5.
AI modelIn Settings
DeepSeek’s model in Hazelchat
DeepSeek V4.1 Flash
Open model, thorough. An open model that handles detailed questions well, including ones that draw on several parts of your content.
DeepSeek’s open model reaches Hazelchat through OpenRouter, served by hosting providers on routes that keep no data, not by DeepSeek’s own service. You need no DeepSeek account and no key.
Choose the model
-
In Hazelchat
Pick it in Settings
Open Settings and find AI model. Claude Sonnet 5.5 and GPT-5.4 Mini show first; the others are under More models.
-
In Hazelchat
Try it on your own questions
On Test agent, the AI model line switches the model and asks your last test question again, so you can compare replies on your own content.
-
In Hazelchat
Publish
A change reaches visitors when you publish your agent. Until then they keep the model they had.
What stays the same with any model
Only your content
Every model answers only from the pages, documents and written answers you approved, with the sources under every answer.
Checked before it is shown
Numbers, dates, links and product names in a reply are checked against the quoted excerpts. If one is not there, the visitor sees the excerpts instead.
One answer counts as one
Your monthly allowance is the same with every model. There are no credits per model and no surcharge.
No data kept by the model
Every model is reached through OpenRouter on routes that keep no data, with data collection switched off.
A $0.008 ceiling per answer
A model whose list price passes that line leaves the list until it is back under it. Answers then continue with the next model, and Settings tells you.
Some replies use no model
Exact matches to your written Q&A are answered without a model, and so are some questions your content clearly does not cover.
All eight models on the list
The picker shows each model with one line about what it is good for.
| Model | Lab | What it is good for |
|---|---|---|
| Claude Sonnet 5.5 (recommended) | Anthropic | Fast and accurate |
| GPT-5.4 Mini | OpenAI | Quick answers |
| Claude Sonnet 5 | Anthropic | Earlier Sonnet, steady |
| DeepSeek V4.1 Flash | DeepSeek | Open model, thorough |
| Mistral Small 4 | Mistral | European, quick for simple questions |
| Qwen3.8 2.4T A95B | Qwen | Open model, detailed answers |
| GLM 5.3 | Z.ai | Open model, detailed answers |
| GPT-6.1 Sol | OpenAI | Thinks longer before answering |
The other model pages: Claude, OpenAI GPT and Mistral.
Questions
Does my visitors’ data go to DeepSeek in China?
Requests go through OpenRouter to hosting providers that serve the open model on routes without data retention, not to DeepSeek’s own app or API. The data processing agreement lists who processes what.
When is DeepSeek a good choice?
When visitors ask detailed questions that draw on several parts of your content, such as comparing options or checking conditions.
Which plan do I need?
Any paid plan. Free answers with the recommended model, Claude Sonnet 5.5.
Try the models on your own content.
Start free with the recommended model. A paid plan, with a seven-day trial, opens every model on the list. See pricing.