# Whats your current go-to LLM model? **URL:** \ **Category:** Agents\ **Tags:** llm\ **Created:** [April 23, 2026, 4:38pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941 "2026-04-23T16:38:24Z")\ **Posts on this page:** 9\ **Page:** 1
**Author:** ![Sameer\_Vohra](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/sameer_vohra/32/961_2.png) [@Sameer\_Vohra](https://community.livekit.io/u/Sameer_Vohra)\ **Post date:** [April 23, 2026, 4:38pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/1 "2026-04-23T16:38:24Z")
The big labs are constantly releasing newer models and new providers keep showing up 🙂 Its hard to keep up. So, post your go-to model(s) and providers. If there are \> 10 responses, I’ll maintain a summary of the results for everyone. ---
**Author:** ![Sameer\_Vohra](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/sameer_vohra/32/961_2.png) [@Sameer\_Vohra](https://community.livekit.io/u/Sameer_Vohra)\ **Post date:** [April 23, 2026, 4:39pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/2 "2026-04-23T16:39:42Z")
Currently its `gpt-5-chat-latest` but I just saw OpenAI’s deprecation notice for it. ---
**Author:** ![Raghu\_Udiyar](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/raghu_udiyar/32/334_2.png) [@Raghu\_Udiyar](https://community.livekit.io/u/Raghu_Udiyar)\ **Post date:** [April 29, 2026, 7:36am UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/3 "2026-04-29T07:36:04Z")
The model choice will vary a lot depending on the use case and requirements - and as you mentioned, new models keep getting released. And usually there’s a mix of models in the same agent for different use cases to optimise on cost and latency. ---
**Author:** ![Laurent\_Bo](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/laurent_bo/32/1180_2.png) [@Laurent\_Bo](https://community.livekit.io/u/Laurent_Bo)\ **Post date:** [April 29, 2026, 12:53pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/4 "2026-04-29T12:53:50Z")
do you have examples @Raghu_Udiyar ? like which models to use for which context? curious to understand better the logic behind thx ---
**Author:** ![Sameer\_Vohra](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/sameer_vohra/32/961_2.png) [@Sameer\_Vohra](https://community.livekit.io/u/Sameer_Vohra)\ **Post date:** [April 29, 2026, 1:27pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/5 "2026-04-29T13:27:24Z")
Thats totally fair. `gpt-5-chat-latest` has been good for conversations that are 2-10mins and a good balance of instruction following vs latency using about 5-15K context. ---
**Author:** ![RabbaniF22](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/rabbanif22/32/1002_2.png) [@RabbaniF22](https://community.livekit.io/u/RabbaniF22)\ **Post date:** [April 30, 2026, 5:19am UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/6 "2026-04-30T05:19:43Z")
I use gpt 4o-mini. cheaper and faster. ---
**Author:** ![Takashi\_Futada](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/takashi_futada/32/530_2.png) [@Takashi\_Futada](https://community.livekit.io/u/Takashi_Futada)\ **Post date:** [April 30, 2026, 6:47am UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/7 "2026-04-30T06:47:02Z")
RealtimeAPI 1.5 on Azure. it works with Japanese. Gemini Flash Live could be okay but some gotchas on tool calling for now. ---
**Author:** ![Sameer\_Vohra](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/sameer_vohra/32/961_2.png) [@Sameer\_Vohra](https://community.livekit.io/u/Sameer_Vohra)\ **Post date:** [April 30, 2026, 1:54pm UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/8 "2026-04-30T13:54:43Z")
Interesting - both of those make sense. Has the instruction following and tool calling been good enough that you haven’t needed to look at newer models ? ---
**Author:** ![RabbaniF22](https://sea1.discourse-cdn.com/flex001/user_avatar/community.livekit.io/rabbanif22/32/1002_2.png) [@RabbaniF22](https://community.livekit.io/u/RabbaniF22)\ **Post date:** [May 5, 2026, 11:45am UTC](https://community.livekit.io/t/whats-your-current-go-to-llm-model/941/9 "2026-05-05T11:45:39Z")
I cannot say it does follow instructions all the time. I can say it works 90% of the time. But ours is a complex scenario. but for simple to medium tasks the model should do just good. Tool calling is good.