Source
State of Open Models: Summer 2026 Observations
Hugging Face's Summer 2026 report analyzes open AI model trends from January to August 2026. It details growth in repositories and the shift toward large-scale Chinese frontier models.
Why it matters
Increased optimization and permissive licensing allow developers to run frontier-scale models on consumer hardware, lowering barriers to high-performance AI.
The details
- Chinese labs released models between 754B and 2.78 trillion parameters in 2026.
- Qwen has become a foundation with 151,448 derivative repositories on the Hub.
- Autonomous agents now represent a significant portion of Hugging Face Hub traffic.
Show entities and relationshipsHide entities and relationships
In this article
Products
Companies
Topics
Organizations
Countries
Key connections
Hugging Face owns lerobot
Hugging Face maintains the open-source lerobot library and repository.
Hugging Face owns Hugging Face Hub
Hugging Face operates the Hugging Face Hub platform.
Hugging Face owns Spaces
Hugging Face operates Spaces for hosting AI demos and agent endpoints.
Hugging Face owns hf_fs
Hugging Face developed the hf_fs MCP tool.
Hugging Face owns huggingface_hub
Hugging Face maintains the huggingface_hub client library.
Hugging Face owns hf CLI
Hugging Face maintains the hf CLI developer tool.
Show 47 more connectionsShow fewer connections
Hugging Face owns transformers
Hugging Face maintains the transformers library.
Hugging Face owns peft
Hugging Face maintains the peft parameter-efficient fine-tuning library.
Hugging Face owns diffusers
Hugging Face maintains the diffusers library.
Thinking Machines Lab owns Inkling
Thinking Machines Lab developed the Inkling 952B model.
Arcee AI owns Trinity-Large
Arcee AI developed and released the Trinity-Large model.
Meta owns Muse Glimmer
Meta developed and released the Muse Glimmer open model.
Meta develops the Llama family of open models.
IBM develops the Granite family of enterprise AI models.
Alibaba develops the Qwen family of open-weight models.
Alibaba owns Qwen 3.8 Max
Alibaba released the Qwen 3.8 Max 2.4-trillion parameter model.
Moonshot develops the Kimi K3 open model.
DeepSeek owns DeepSeek-V4-Flash
DeepSeek develops the DeepSeek-V4-Flash open model.
Apple develops the MLX machine learning framework for Apple silicon.
The ggml team creates and maintains the llama.cpp project.
ggml team is a member of Hugging Face
The ggml team joined Hugging Face in February 2026.
Agentic AI Foundation is a member of Linux Foundation
Agentic AI Foundation operates under the Linux Foundation.
Agentic AI Foundation is related to Model Context Protocol
The Model Context Protocol moved into the Agentic AI Foundation.
hf_fs is built with Model Context Protocol
hf_fs is a tool designed for Model Context Protocol servers.
llama.cpp uses the GGUF file format for local inference.
Qwen models are distributed in GGUF quantized builds.
llama.cpp is related to Local Inference
llama.cpp is a key software tool enabling local inference on consumer hardware.
Kimi K3 is built with Mixture of Experts
Kimi K3 is a mixture-of-experts model architecture.
Claude Code uses Hugging Face Hub
Claude Code connects to and interacts with Hugging Face Hub.
Codex uses Hugging Face Hub
Codex connects to and interacts with Hugging Face Hub.
all-MiniLM-L6-v2 is related to Hugging Face Hub
all-MiniLM-L6-v2 is one of the most downloaded models on Hugging Face Hub.
Hugging Face uses GLM-5.2
Hugging Face used GLM-5.2 to analyze an autonomous agent intrusion.
NVIDIA and AMD are leading hardware companies competing in open model releases.
Qwen and Llama compete as foundational base models for downstream derivatives.
Moonshot competes with MiniMax
Moonshot and MiniMax compete in releasing frontier-only large open models.
Google and Meta compete in open model releases and derivatives.
IBM is located in United States
IBM is based in the United States.
LiquidAI is located in United States
LiquidAI is based in the United States.
Thinking Machines Lab is located in United States
Thinking Machines Lab is based in the United States.
Arcee AI is located in United States
Arcee AI is based in the United States.
Linux Foundation is located in United States
The Linux Foundation is based in the United States.
Moonshot is an AI lab based in China.
MiniMax is an AI lab based in China.
Xiaomi is based in China.
Z.ai is an AI lab based in China.
Tencent is based in China.
Alibaba is based in China.
Ant Group is based in China.
Meituan is based in China.
DeepSeek is an AI lab based in China.
AMD is located in United States
AMD is headquartered in Santa Clara, California, United States.
Unsloth uses Quantization
Unsloth publishes quantized and fine-tuning-ready model builds.
Unsloth publishes builds extending the Qwen model ecosystem.
Related events
MCP Transitions to Linux Foundation Agentic AI Foundation
Hugging Face Releases Summer 2026 State of Open Models Report
ggml Team Joins Hugging Face
Autonomous Agent Conducts Intrusion on Hugging Face Infrastructure
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.