Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-16 · high
Claude Code alone drives 44.4% of that agent traffic, per Hugging Face's August 14 open-models report.
Why now: Hugging Face published the Summer 2026 report on August 14, covering Hub usage through July 2026.
Agents overtook humans as Hugging Face's primary Hub users by July 2026, per the company's Summer 2026 State of Open Models report. Claude Code alone generated 44.4% of that agent traffic.
That flips the design target for anyone publishing model weights on the Hub. A README and a file name now get read by a parser more often than by a person scrolling past it. Metadata, quant naming, and file structure decide whether a tool call finds a model, not whether a human likes the page.
The report's download numbers back up the shift toward machine-readable adoption over headline attention. Qwen pulls 39.6M monthly GGUF downloads against Gemma's 20.8M and Llama's 7.5M, and holds 151,448 derivative repos, 2.6 times Meta's total footprint. Only one repository appears on both the top-25-by-downloads list and the top-25-by-likes list. Attention and adoption measure different things now, and adoption is the one agents act on.
Models under 1B parameters take 83% of all-time downloads. Everything above 100B takes 1%. The report does confirm the trillion-parameter local story is real. July's snapshot lists GGUF builds of DeepSeek-V4-Flash at about 284B parameters and Kimi-K3 at about 2.8T, and almost nobody pulls them. Public model repos grew from 2.43M to 2.96M over the period, datasets from 711K to 1M, Spaces from 1.00M to 1.44M.
If your product serves model artifacts, your primary client is now a machine that never scrolls.
Each link below shares sources, entities, or timing with this story.
The August 14 report covers January through August 2026: model repos grew from 2.43M to 2.96M, datasets from 711K to 1M, and 85.6% of models have under 200 lifetime downloads (Hugging Face). Chinese labs shipped monthly parameter ceilings of 754B to 2.78T against sub-130B for...
For about a year, "run your agent locally" meant accepting a model that couldn't reliably call a tool twice in a row. That excuse is gone. Meta Superintelligence Labs published Muse Glimmer today: a 29.6B dense causal transformer, 52 layers, 6,656 hidden dim, with a ~1.8B ViT-...
Xiaomi released MiMo-V2.5-Pro, a 1.02 trillion parameter mixture-of-experts model (42B active) with 1M token context, fully MIT licensed. In benchmarks, it achieves 63.8% success on agentic tasks using 40-60% fewer tokens than Claude Opus 4.6 or GPT-5.4 for comparable results....
The leaderboard says first place. The methodology says you should check your own bill. Qwen3.8 Max now ranks first on Artificial Analysis' agentic index, scoring 86.1 on OSWorld-Verified ahead of GPT-5.6 Sol Max at 83.2 and Fable 5 at 85.0, priced at $2.00/M input and $6.00/M...
Created August 24, it holds a trendingScore of 3,967 against second-place GLM-5.3-Flash at 1,376 (Hugging Face). The near-1:1 like-to-download ratio means almost everyone bookmarking it hasn't pulled weights, and the unsloth GGUF conversion at 4,354 downloads is absorbing comp...
The models are good. The license is the real story. Google released Gemma 4 on April 2 with four variants: E2B, E4B, 26B MoE, and 31B Dense. All built on the Gemini 3 architecture. The 31B Dense variant claimed #3 on Arena AI's text leaderboard, beating models 20x its size. Th...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.