Any of the newer Qwen3 reasoning models:
- Qwen/Qwen3-235B-A22B-Thinking-2507
- Qwen/Qwen3-Next-80B-A3B-Thinking
We’re currently using Qwen3 32B, and would like an upgrade in intelligence
GLM 4.6 and DeepSeek V3.1 Terminus are extremely capable models according to artificial analysis
PS: I can’t put more than one embedded image, but GLM 4.6 only wins against DeepSeek V3.1 Terminus in Agentic Index, but DeepSeek has a very low score there. Otherwise, DeepSeek is in front of GLM 4.6 and sometimes even in front of Gemini 2.5 Pro.
See it here: AI Model & API Providers Analysis | Artificial Analysis
We’re considering / testing this one, and generally a newer image model! We haven’t landed on the best one to launch yet though
1 Like
Any plans on adding OCR models? Now I’m using Mistral OCR via API but it would be nice to have Deepseek OCR on Groq
We’re considering an image model, but no plans on OCR specific models right now. Deepseek OCR is interesting, but I think it’s more of a research model on compression than a full production-level OCR model?
Dear Groq Team,
I would like to kindly ask if you could consider adding the Polish large language model Bielik-11B-v2 to the Groq platform. This model, developed by the SpeakLeash project, represents one of the most capable open Polish LLMs currently available and would significantly enhance multilingual support for Groq users and developers.
Thank you for the recommendation! We’ll consider it, but most likely the use case is too small for this model to be deployed on Groq — have you considered using Replicate?
Here’s the model inference endpoint on Replicate: aleksanderobuchowski/bielik-11b-v2.3-instruct | Run with an API on Replicate
pyannote/speaker-diarization-community-1
I would love for this model to be hosted by Groq. It would compliment the Whisper models!
Oh that’s a great model! We’re looking at different kinds of audio models right now, stay tuned!
Awesome. Speaker diarization would be a great addition to the stt models. Renting out a beefy GPU server isn’t feasible atm.
What kind of audio models are you guys looking into? Chatterbox, and VoxCPM are great TTS models!
1 Like
We hear you! It’s something that’s kind of been on the backburner to be honest, but it’s something we really want
GLM 4.6 is great for agentic tool use. Why don’t you guys have those models on here?
1 Like
We’re considering it! It’s pretty good for some use cases but I don’t think it’s like 10x better than a lot of the other open source models.
Minimax-M2 is currently the leading open weight model for agentic tasks. It’s also far smaller than Kimi-K2.
1 Like
Oh interesting, this hasn’t come across my radar at all. I need to play with this one!
K2 is super capable!
GPT-OSS is great for a lot of business use cases like json data or text extraction and tagging, but yeah it falls behind on agentic coding work
I’d love to see this as well
1 Like
Would love to see the new Kimi K2 Thinking! Any timeline for adding it?
3 Likes
We’re looking into it, but no timeline yet. We’ll keep you posted!
1 Like
parakeet 0.6b for ASR!! You guys have whisper currently but parakeet crushes it in accuracy benchmarks. Supposedly faster too. nvidia/parakeet-tdt-0.6b-v2 · Hugging Face