Alternatives to Keras Hub

KcBERT: a free Korean BERT pretrained on 12.5GB of Naver news comments for messy, slang-heavy text.

RobBERT is a family of open pre-trained Dutch language models you fine-tune for classification, regression, and token-tagging tasks.

Browse 2M+ open models, 500k+ datasets and 1M+ Spaces on the Hugging Face machine learning hub

Mistral: open-weight AI models you can self-host, fine-tune and run in-region

Zhipu GLM delivers open-weight GLM-5.3 models with 1M-token context, MaaS APIs, and shipped computer-use agents.

Cohere delivers enterprise RAG and sovereign AI models you can run in your own VPC or on-premises.

KcELECTRA-base-v2022: Korean ELECTRA pretrained on 162M Naver News comments for noisy Korean NLP

ModelScope is Alibaba's open model hub for hosting, fine-tuning, evaluating and demoing Qwen, MiniMax and DeepSeek weights.

Open-weight agentic coding models: Laguna S 2.1 (118B) and on-device Laguna XS 2.1 (33B).

SmallDoge small language models: open SLM weights, training checkpoints, and SLM datasets on Hugging Face

AI21 post-trains specialized models on your proprietary data for frontier-level quality at lower cost per call

PaLM API gives developers safe, scalable access to Google's language models, paired with MakerSuite for prompt prototyping.

Ludwig trains multi-modal and LLM models from a single YAML config file — no training loops, Apache 2 licensed.

Falcon LLM: Apache 2.0 open-weight models for Arabic, Emirati and multimodal AI you self-host

Magic.dev builds frontier code models with ultra-long context to automate software engineering and AI research.

Feyn trains custom AI models on your proprietary data so you own the weights and keep improving in production.

NXAI builds xLSTM memory architectures for edge and industrial time-series work, shipping TiRex-1 and TiRex-2 models on Hugging Face and GitHub.

Open-weight frontier models trained with reinforcement learning for domain knowledge work in finance, law, consulting and healthcare.

Open-source multilingual BERT for 11 Indic languages and Indian-English.

Pay-per-token access to MiniMax M3 and M3.1 Flash language APIs plus H3 video, Speech 2.8 and Music 3.0.

Transformers is Hugging Face's open-source model-definition framework for running, fine-tuning, and serving text, vision, audio, and

Sarvam AI brings Indic-language text to speech, speech-to-text, translation and document parsing to Indian voice agents.

Baichuan-7B: a free, ungated 7B Chinese-English base model you download and fine-tune yourself.

Qwen3.6-27B: Apache 2.0 agentic coding and vision reasoning on your own GPU.

KoBigBird brings long-document Korean NLP up to 4096 tokens with BigBird block sparse attention.

StepFun's API delivers OpenAI-compatible frontier reasoning, vision-language document parsing and a full StepAudio 3 voice stack.

MGM-Omni: open-weight omni-modal LLM demo fusing speech, text, and vision with zero-shot voice cloning.

Open-source add-on to Hugging Face Transformers for training, composing, and sharing parameter-efficient fine-tuning adapters.

Chinese medical LLMs with open weights from 7B to 235B plus the 百小医 AI family-doctor app

Dolly is Databricks' open-source instruction-following LLM you can fine-tune on a single A100 in about 30 minutes.