Hi everyone,
We’d love to hear what kind of models you’d love to see on Groq — from coding models to text-to-speech and speech-to-text, to embeddings and diffusion and other sorts of models.
While we can’t accommodate everyone’s wishes, we’d love to keep the conversation alive here with new model drops, and model benchmarks and performance in this thread!
I”d love to see portuguese text-to-speech
phi would be nice if it was added
We’re exploring tts models at the moment; what’s your favorite Portuguese tts model?
We’re considering the phi series, but honestly that one’s a bit lower on the consideration list than some of the other ones
Considering cost-benefit I’d say that AI products for growing applications. (excelent princing and good quality) or https://murf.ai/ (good price, better quality). There are others, but for heavy usage I think that no one beats those two cost-benefit wise.
One or two more vision models to compete with the Llama 4 ones that you guys have would be great. Qwen2.5 VL would be good!
we’re currently exploring Qwen2.5 VL! I’ve heard great things about it too, and really good point about getting more vision models
We are on the look-out for an elastic, scalable environment for our speech-to-text inference workloads. We use the NB-Whisper models - based on Whisper from OpenAI and trained further on Norwegian speech data - from NbAiLab ( NB-Whisper - a NbAiLab Collection ).
We are currently serving these from a dedicated H100 environment. Due to client consumption growth we are looing for a more compute-efficient solution, and we would love to try out groq. But we are entirely dependent on using these Norwegian-specific models, and would like to host them from the Nordics (read; Helsinki).
That’s really interesting, I didn’t know about a Norwegian specific Whisper! This is probably too niche for us to host for the immediate future, but as a fellow Nordic (Swede!) it’s really cool to see a Norwegian-tuned Whisper. Probably every language (and dialect) will have their own speech-to-text model in the future!
I brought it up before in another thread but for the sake of visibility, I’ll post it again here.
I’m desperately missing EU compliant VLMs (and non-reasoning LLMs) and I’m also desperately missing capable SLMs with very low latency and high throughput. I’m currently still on Llama3.1-8B but this is apparently slower then the Llama4 series and it also is not producing json-structured output reliably.
I therefore suggest the Mistral and/or Gemma3 family of models. A combination of Mistral Small 3 and Mistral Medium 3 should be rock solid.
Any deepseek model.
Or another one for creative writing, that doesn’t “feel” like a robot is writing.
I am looking for a model to integrate with my Kilo Code environment. I’ve search for Devstral but found nothing for my purpose.
@Per is there an open model for creative writing that you really like?
@TurkerT have you tried Kimi K2 yet as a coding model? I haven’t used Kilo Code yet, but K2 is pretty good!
Would love to see Qwen3 235B Instruct or Deepseek V3.2!
i would like to see these models, these are alot but these are good ones- deepseek r1, parakeet, glm 4.5, deepseek 3.1, stable diffusion 3.5.
There is a separate thread for **Z.AI: GLM 4.5, but 4.6 is out now and showing excellent stats.
Would love to see it hosted on Groq!**
I would love to see open source model specially medical related like : MedGemma and more.
Mistral-OCR 
=> powerful for unstructured data extraction workflows 