Skip to content
AI Compute Radar
IndexCollected 3 h ago

Maker board

Who leads,by the numbers.

15 makers build the 49 open models this site tracks. This board says which of them hold the top slots by popularity today, how their models' interest moved over seven days and where the downloads go — sums and means of the same measured readings as the front page, never an opinion.

How popularity is scoredThe live departures board

How to read it

  • Top-10 slots: how many of the maker's models sit in the top 10 by popularity right now — the ordering criterion.
  • Mean Heat: the average score of the maker's scored models; the 7-day change is the average move of the models that had a reading a week ago.
  • 30-day downloads: the Hub's rolling count per tracked repository, summed per maker — a repository count, not a user count.
  • Newest release: the youngest repository creation date among the maker's tracked models, with the releases of the last 90 days counted next to it.
  • Picks: published Picks of the week won by the maker's models.
  • Only tracked models count. A maker with one tracked model is judged on that one model.

The standings today

Order: most models in the popularity top 10 first, ties by the best-placed model, then by 30-day downloads. A maker is the creator named on the model page. Downloads are the Hub's 30-day counts of the tracked repositories, summed; the 7-day change is weighted by those counts.

The standings today
MakerTop-10 slotsBest-placed modelMean Heat30-day downloads
1Alibaba Qwen11 of 11 scored3Qwen3-Coder 30B-A3B#3 · Heat 7554▼ −2.5 over 11 models, 7 days53.4 M▼ −6 % in 7 days
2OpenAI2 of 2 scored2gpt-oss 120B#1 · Heat 7871▲ +5.5 over 2 models, 7 days8.0 M▼ −2 % in 7 days
3Zhipu AI5 of 5 scored2GLM-5.3-Flash#4 · Heat 7153▼ −3.8 over 5 models, 7 days10.6 M▼ −3 % in 7 days
4Google5 of 5 scored1Gemma 4 E4B#2 · Heat 7755▼ −0.6 over 5 models, 7 days17.3 M▼ −6 % in 7 days
5Liquid AI5 of 5 scored1LFM3 1.7B#5 · Heat 7037▼ −2.2 over 5 models, 7 days299 K▼ −2 % in 7 days
6Meta3 of 3 scored1Llama 3.1 8B#8 · Heat 6662▼ −2.3 over 3 models, 7 days5.3 M▼ −2 % in 7 days
7Ornith AI4 of 4 scored0Ornith 1.5 35B-A3B#12 · Heat 6452▼ −0.3 over 4 models, 7 days1.6 M▲ +19 % in 7 days
8Kwaipilot1 of 1 scored0KAT-Coder V2.5#13 · Heat 6363▲ +17 over its one model, 7 days73 K▲ +354 % in 7 days
9DeepSeek3 of 3 scored0DeepSeek-V4-Flash#18 · Heat 5949▲ +4.7 over 3 models, 7 days3.5 M▼ −17 % in 7 days
10IBM3 of 3 scored0Granite 4.2 8B#28 · Heat 4741▼ −6.7 over 3 models, 7 days134 K▼ −4 % in 7 days
11NVIDIA1 of 1 scored0Nemotron Nano 9B v2#30 · Heat 4444▲ +10 over its one model, 7 days172 K▲ +48 % in 7 days
12Microsoft1 of 1 scored0Phi-4 Mini#34 · Heat 4141▲ +12 over its one model, 7 days305 K▼ −8 % in 7 days
13InclusionAI2 of 2 scored0Ling 3.0 Tiny#37 · Heat 4038▲ +9 over 2 models, 7 days22 K▼ −5 % in 7 days
14Mistral AI2 of 2 scored0Mistral Small 3.2#38 · Heat 3726▲ +6 over 2 models, 7 days134 K▲ +1 % in 7 days
15Tencent1 of 1 scored0Hy3#44 · Heat 3131▲ +1 over its one model, 7 days935 K▼ −13 % in 7 days

Reading the maker board

As of today's readingsOn November 19, 2026 the board counts 15 makers behind the 49 tracked models. Alibaba Qwen holds the most top-ten slots, 3. The highest mean popularity score across a maker's scored models is OpenAI's, 71. The largest seven-day rise in mean Heat is Kwaipilot's, +17.0 points over one model. The largest seven-day fall is IBM's, −6.7 points over 3 models.

The board counts, it does not rank by opinion. For every maker it sums and averages the same readings the front page shows for each model: how many of its models hold a slot in the top ten by Heat Score, the mean score of its scored models, the mean move of those scores over seven days, the sum of its models' downloads over thirty days and how that sum moved, and when its newest tracked model was released. A maker with one hot model and a maker with five warm ones can land next to each other; the columns say which is which.

Two things keep the board honest. First, only tracked models count, and the catalogue is a curated set of open models with measured files, not every repository on the Hub; a maker's rank here says where it stands among those, nothing more. Second, every column is reconstructible from the model pages: the board stores no number of its own, so when a model's Heat Score changes, the board changes with it the next time it renders.

Read the seven-day column with the model count next to it. A mean move over one model is that model's move; a mean over eight is a trend across a lineup. The pick-of-the-week marker on a maker's row says that one of its models was chosen by the pick rule; it is a fact about attention in that week, not an endorsement of the maker.

Their models

Every tracked model by maker, in popularity order. Rank and score are today's; a model without a score is still in its first week of history.