## patner resources

### OpenBMB

The MiniCPM family of models are tiny and highly capable models for a variety of usecases spanning many test, vision, and audio.

Includes MiniCPM 1B (ultra-fast local text), MiniCPM 4.18B (deep reasoning), MiniCPM-V 4.6 (1.3B param vision/document assistant), and MiniCPM-o 4.5 (world's first full-duplex omni-model for natural text/audio/vision).

OpenBMB have provided free API access to use these models.

API endpoints:
- **MiniCPM4.1-8B**: http://35.203.155.71:8001
- **MiniCPM-V-4.5**: http://35.203.155.71:8002
- **MiniCPM-V-4.6**: http://35.203.155.71:8003
- **MiniCPM-V-4.6-Thinking**: http://35.203.155.71:8004\

Authorization:
```
Bearer sk-minicpm-V8bcD-YTAMxECagaKOnbwTCN69|IN2LhSezGOgq2Ues
```

These models can also be run locally on most hardware via llama.cpp or transformers.

#### Recommended MiniCPM Models

The following models are recommended for participants building with MiniCPM models. Participants may also use other rule-compliant MiniCPM series models as long as they comply with the hackathon’s model-size requirements.
Model Best For Link

OpenBMB Model Collections
Central entry point for all OpenBMB / MiniCPM
models on Hugging Face
https://huggingface.co/openbmb/collections

MiniCPM-V 4.6 

Image understanding, video understanding, OCR, document understanding, multimodal apps
https://huggingface.co/openbmb/MiniCPM-V-4.6

MiniCPM-o 4.5 

Omni-modal interaction, voice + vision +
language, real-time multimodal experiences
https://huggingface.co/openbmb/MiniCPM-o-4_5

MiniCPM-V 4.5 

Strong multimodal understanding,
image/video use cases, visual reasoning
https://huggingface.co/openbmb/MiniCPM-V-4_5

MiniCPM5-1B 

Lightweight text generation, local-first apps,
on-device assistants, small LLM use cases
https://huggingface.co/openbmb/MiniCPM5-1B

MiniCPM4.1-8B 

Text reasoning, efficient inference, hybrid
reasoning, local reasoning apps
https://huggingface.co/openbmb/MiniCPM4.1-8B

VoxCPM2 

Text-to-speech, voice generation, creative voice
design, voice-enabled AI apps
https://huggingface.co/openbmb/VoxCPM2


#### Model Selection Guide

If you want to build… Recommended Model

Image understanding app: MiniCPM-V 4.6
OCR / document assistant: MiniCPM-V 4.6
Video understanding demo: MiniCPM-V 4.6 or MiniCPM-V 4.5
Omni-modal assistant: MiniCPM-o 4.5
Voice + vision interaction: MiniCPM-o 4.5
Lightweight text assistant: MiniCPM5-1B
Local-first text app: MiniCPM5-1B
Text reasoning / problem-solving app: MiniCPM4.1-8B
Efficient local reasoning app: MiniCPM4.1-8B
TTS / voice generation app: VoxCPM2
Creative voice experience: VoxCPM2
Mobile / offline multimodal app: MiniCPM-V Apps + MiniCPM-V series

#### Developer Support Documents

MiniCPM-V / MiniCPM-o Cookbook — GitHub
Example projects, usage recipes, application ideas, and developer guides
https://github.com/OpenSQZ/MiniCPM-V-CookBook

MiniCPM-V / MiniCPM-o Cookbook — Website
Web version of the cookbook; easier for participants to browse
https://opensqz.github.io/MiniCPM-V-CookBook/site/en/index.html

MiniCPM-V Apps 
Mobile app deployment examples and offline app resources
https://github.com/OpenBMB/


- Website: https://www.openbmb.cn/en/home
- Huggingface: https://huggingface.co/openbmb
- MiniCPM collection: https://huggingface.co/openbmb/collections
- GitHub: https://github.com/OpenBMB
- Twitter: https://x.com/OpenBMB
- Youtube tutorials: https://www.youtube.com/channel/UCyDtv4OTmeZ1OYrCHymYGow?utm_source=chatgpt.com\
- Discord support:
	- @zzhonglol2_50531
	- @tc_mb
	- @yangzhizheng

### Black Forest Labs

Models: FLUX.2 Klein base and distilled versions (available in 4B and 9B parameter sizes). Excellent for text-to-image generation and precise image editing workflows.

- Website:
- huggingface:
- Twitter:
- Github: 
- Docs: docs.bfl.ai
- Guide: hf.co/blog/black-forest-labs/flux-2-klein-lora
- Starter Space: hf.co/spaces/stephenbtl/klein-build-small-starter
- Model: FLUX.2-klein-base-4B • Apache 2.0
- Trainer: github.com/ostris/ai-toolkit
- Discord support: @stephen.btl

### OpenAI Codex

The powerful Codex coding agent (GPT-5.5) with built-in ecosystem plugins (GitHub, Figma, Hugging Face integrations), SSH connectivity to VMs, in-app UI browser checks, and the new automatic "Goal Mode" which can carry out multi-hour developer tasks unprompted.

- Website:
- Codex docs: 
- Huggingface:
- Twitter:
- Github:

### NVIDIA

Models: Nemotron 3 Nano (30B total, 3B active MoE), Nemotron 3 Nano 4B (edge), Nemotron 3 Nano Omni (multimodal), Nemotron Cascade (math/code), Nemotron 3 ASR (speech recognition), Nemotron Parse (< 1B parameter document extraction), and Nemotron Embed VL (vision-language embeddings).

- Website:
- Huggingface:
- Nemotron models:
- Twitter:
- Github:
- Nemotron 3 Nano usage guide: https://github.com/NVIDIA-NeMo/Nemotron/tree/main/use-case-examples/Simple%20Nemotron-3-Nano%20Usage%20Example
- Nemotron 3 Ultra: https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/

### Modal

Infrastructure for all AI workloads.

#### inference

Deploy and scale inference for LLMs, audio, image/video generation.

#### training

Fine-tune open-source models on single or multi-node clusters instantly.

#### batch

Parallelize massive jobs with Modal's hyper-elastic compute infrastructure.

#### sandboxes

Programmatically scale ephemeral environments for running untrusted code.

Tools: Elastic, serverless compute infrastructure ideal for running fast inference, fine-tuning scripts in ~300 lines of code, parallel hyperparameter sweeps, and deploying coding agents within remote sandboxed environments.

- Website:
- Docs: 
- Huggingface:
- Twitter:
- Github:
- Discord support: @felicia_modal
- Slack: modal.com/slack

### JetBrains

Models: Mellum 2 (12B parameter mixture of experts). Optimized for coding and high-throughput logic requests. Available in dedicated "Thinking" (reasoning-heavy) and "Instruct" (blazingly fast) configurations.

What is Mellum2?

• Family of open-source language models from JetBrains
• Built for coding and language tasks
• Optimized for low-latency, high-throughput inference
• Apache 2.0 licensed
• Deploy locally or in the cloud

What can you build?
• Al coding assistants
• RAG applications
• Intelligent routing between models
• Code analysis and developer tools
• Real-time chat and automation

- Website:
- Huggingface:
- Twitter:
- Github:
- GGUF Collection: https://huggingface.co/collections/JetBrains/mellum2-gguf 
- Quickstart: https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking-GGUF-Q4_K_M
- Discord support:
	- @nikitapavlichenko
	- @vano006503
	- @aral_dm

### Cohere Labs

Cohere Transcribe is a 2B parameter fast ASR model covering 14 languages. The Aya 3.3B series (Base, Global, Earth, Fire, and Water variants) are tiny text generation models, each targeting and optimized for distinct geographical language families and low-resource languages.

- cohere transcribe: https://huggingface.co/CohereLabs/cohere-transcribe-03-2026
- tiny-aya-global: Best balance across languages and regions - https://huggingface.co/CohereLabs/tiny-aya-global-GGUF
- tiny-aya-water: European and Asia-Pacific languages - https://huggingface.co/CohereLabs/tiny-aya-water-GGUF
- tiny-aya-fire: South Asian languages - https://huggingface.co/CohereLabs/tiny-aya-fire-GGUF
- tiny-aya-earth: West Asian and African languages - https://huggingface.co/CohereLabs/tiny-aya-earth-GGUF

Together, they are a good fit for local multilingual assistants, voice interfaces, accessibility tools, offline translation helpers, and small apps for real people.

Resources 

- Cohere transcribe release blog: https://huggingface.co/blog/CohereLabs/cohere-transcribe-03-2026-release
- Tiny Aya example space: https://huggingface.co/spaces/ysharma/tiny-aya-build-small-sample
- Tiny Aya and Cohere transcribe full guide: https://huggingface.co/blog/CohereLabs/build-small-hackathon-with-cohere-models

- Website:
- Huggingface:
- Twitter:
- Github:
- Discord support:\
	- @madeline_smith
	- @alejandro_81346
	- @julianmack_43074