# I'd like to know which LLMs everyone is using with the Discourse AI plugin

**URL:** https://meta.discourse.org/t/discourse-ai/412905
**Category:** UX
**Tags:** ai
**Created:** [September 21, 2026, 10:29am UTC](https://meta.discourse.org/t/discourse-ai/412905 "2026-09-21T10:29:08Z")
**Posts on this page:** 5
**Page:** 1

<div class="post-metadata">

### Author: ![sniper756](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/sniper756/32/545002_2.png) [@sniper756](https://meta.discourse.org/u/sniper756)
#### Post date: [September 21, 2026, 10:29am UTC](https://meta.discourse.org/t/discourse-ai/412905/1 "2026-09-21T10:29:08Z")

</div>

I found that the custom DeepSeek v4.1 flash model on the official DeepSeek website is not working, and there are issues with the Qwen large model’s AI search.

I’d like to know what everyone else is using. Are all the Discourse AI features available?

Thanks

---

<div class="post-metadata">

### Author: ![RGJ](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/rgj/32/523185_2.png) [@RGJ](https://meta.discourse.org/u/RGJ)
#### Post date: [September 21, 2026, 11:03am UTC](https://meta.discourse.org/t/discourse-ai/412905/2 "2026-09-21T11:03:11Z")

</div>

> [@sniper756](#):
>
> DeepSeek v4.1 flash model on the official DeepSeek website is not working

It is? And it’s working pretty well too.

What does “not working” mean?

---

<div class="post-metadata">

### Author: ![sniper756](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/sniper756/32/545002_2.png) [@sniper756](https://meta.discourse.org/u/sniper756)
#### Post date: [September 21, 2026, 1:22pm UTC](https://meta.discourse.org/t/discourse-ai/412905/3 "2026-09-21T13:22:22Z")

</div>

Successfully configured DeepSeek in the admin panel:

 ![image](https://global.discourse-cdn.com/meta/original/4X/f/4/4/f44e24e82408acf12977b60164b01999638cdbe0.png)

 ![image](https://global.discourse-cdn.com/meta/original/4X/8/f/6/8f6943b20378c0e75cba83d908b2d617cf374d97.png)

 ![image](https://global.discourse-cdn.com/meta/original/4X/2/0/c/20cf3fb8fed52f3d9d55a7e8ecb8cd6515524766.png)

1. Unable to use AI to summarize topics

```plaintext
{
  "error": {
    "message": "This response_format type is unavailable now",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_request_error"
  }
}

```

1. Ask AI cannot complete this search.

 ![image](https://global.discourse-cdn.com/meta/original/4X/a/f/4/af43f0881ed4c6db5f98330560e0ec0fd33fb344.png)

---

<div class="post-metadata">

### Author: ![satonotdead](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/satonotdead/32/447830_2.png) [@satonotdead](https://meta.discourse.org/u/satonotdead)
#### Post date: [September 21, 2026, 2:15pm UTC](https://meta.discourse.org/t/discourse-ai/412905/4 "2026-09-21T14:15:54Z")

</div>

I can confirm that Deepseek and also GLM models needs some custom hooks for Discourse AI pairing.

I did it through LiteLLM as a proxy for the final endpoints but a core implementation would be appreciated.

---

<div class="post-metadata">

### Author: ![Orioni](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/orioni/32/542679_2.png) [@Orioni](https://meta.discourse.org/u/Orioni)
#### Post date: [September 21, 2026, 2:48pm UTC](https://meta.discourse.org/t/discourse-ai/412905/5 "2026-09-21T14:48:07Z")

</div>

Gemini 2.5 Flash Lite.

Previously I had briefly used 3.1 Flash Lite. But the quality remained about the same while price went up, so I switched back to the cheaper 2.5FL.

3.1FL cost only about €0,10/ day. 2.5FL took that further down to €0,01/day.