Parakeet is great! I’ve been trying it on personal projects, but I’m not sure Groq as a company wants to adopt it, as it doesn’t fulfill some of the language requirements we have. We’re currently still scouting for a good ASR (e.g. the new Meta one that just came out…!)
I would love to see Claude Sonnet. It’s been the best one I’ve used so far for coding.
1 Like
Me too! Unfortunately that’s up to the Anthropic team
Minimax2 LLM- It is by far the best!
1 Like
Here’s a polished and more persuasive version of your message, along with supporting arguments based on recent developments:
-–
Subject: Request to Add GLM-4.6 and MiniMax M2 to Groq Model Portfolio
Hi Groq Team —
I hope you’re doing well. I’d like to request that you consider adding GLM-4.6 and MiniMax M2 to your supported models. Here are my reasons:
-–
Why GLM-4.6
Open-source and powerful: GLM-4.6 is released under a permissive license (MIT), so it’s fully deployable.
Massive context window: It supports up to 200K tokens, which is a big jump and very useful for complex, long-horizon tasks or large codebases.
Excellent coding performance: On CC-Bench, it reportedly has a ~48.6% win rate against Claude Sonnet 4 in real-world coding tasks.
Improved efficiency: Compared to GLM-4.5, the new version uses ~15% fewer tokens for many tasks.
Strong reasoning + agent abilities: Better reasoning capabilities and native tool use during inference make it well-suited for agent workflows.
Deployment flexibility: The model weights are publicly available (Hugging Face, ModelScope), and local inference is supported via vLLM and SGLang.
-–
Why MiniMax M2
Designed for code + agents: MiniMax M2 is explicitly optimized for coding workflows and agentic tasks.
Efficient architecture: It uses a Mixture-of-Experts (MoE) design with 230B total parameters but only 10B active, which gives a good balance of performance and efficiency.
Huge context: Supports 200k token context, which is very helpful for multi-file projects, long reasoning chains, and tool-heavy agents.
Cost and speed advantages: According to their launch, its inference is about 2x faster than some proprietary models and cost is very competitive.
Open source & commercial use: Fully open-sourced, and can be used in production.
Proven benchmarks: Strong on reasoning, coding, and multi-step agent evaluation.
-–
Why This Matters for Groq
Supporting these models would expand your model coverage to include top-tier open-source coding + agent models, making Groq more attractive to developers working in those workflows.
These models’ efficiency (especially MoE in both and quantization / large context in GLM-4.6) make them potentially well-suited for cost-effective inference on high-performance hardware like yours.
By supporting them early, Groq can position itself as a go-to accelerator for cutting-edge open-weight models.
-–
I believe adding both GLM-4.6 and MiniMax M2 would be a strong strategic win. Thanks for considering this — I’d be happy to help with any additional details, benchmarks, or use-case discussions.
Thanks for the strong post! We’re still evaluating these internally
Could you consider adding gpt-oss-120b-safeguard? Not sure why 20b was added but 120b was not. Thanks!
an ocr model, like mistral-ocr or gemma-3-27b
also what about kimi k2 Thinking when are you adding that
1 Like
More TTS models.
PlayAI is shutting down very soon and we need replacements.
Ideally there would be some choices of expensive and cheap models.
Need super fast TTS for real-time interaction.
Waiting for kimi- thinking 
please DeepSeek-V3.2
Mistral 3 models launched today would be great to see! I have been using the open source model launched by OpenAI on Groq, and it seems that Mistral 3 models are better, so I’m keen to try that when available.
Oh yeah, what are you excited to use Mistral for? It’s pretty cool to see Mistral 3 Large as an open weights model
Seconding an OCR model like olmOCR 2
1 Like
would you say something like olmOCR is much better at OCRing that say the state of the art general visual language model out right now?
Kimi K2 thinking! Take my money!
1 Like
Kimi K2 Thinking; it’s been so long ;’(
deepseek r1, glm, tts japanase
DeepSeek Coder V2 is insane. Im creating a agent coding project powered by Groq API (yessir im n1 glazer) and would really really want DeepSeek Coder V2
Please add Kokoro TTS!! It’d be an amazing fit for your platform. Good quality, very fast, and only 82M params. @yawnxyz
1 Like