Open Source in Late August: A Forecasting Model, Torrent-Shipped Weights, and Video With Sound

Open-source AI releases moved past chat and code in late August. Google’s TimesFM 3.0 brings foundation-model forecasting and ranks first on three benchmark suites, DeepSeek opened its first V4 model with vision (ApexBench up to 36.5%), and modelregistry.io started shipping weights over torrents.
artificial-intelligence
data-science
Author

Kabui, Charles

Published

2026-09-08

Keywords

time-series-foundation-model, open-weights, torrent-distribution, video-with-audio, multimodal-vision

Download as PDF

Open-source releases in late August moved past chat and code. Google’s TimesFM 3.0 is a time-series foundation model that forecasts several related series at once and handles extra inputs like price or weather, ranking first on the fev-bench, TIME, and GIFT-Eval suites; the code is Apache-2.0 but the 3.0 weights are non-commercial. DeepSeek-V4-Flash-Vision-Exp is DeepSeek’s first V4 model with vision (MIT), lifting ApexBench to 36.5% from 26.2% while holding its text-agent scores. MiniMax-H3 makes up to 15 seconds of video with native stereo audio from a 33B model. modelregistry.io serves open weights as HTTPS torrent files with live seed and peer counts, and Chrome DevTools MCP added PWA automation and queryable heap snapshots for coding agents.

A data team can test a foundation forecaster against its own numbers before training per-series models, with TimesFM 3.0 off limits commercially. A developer can run an agent that reads a screen, or generate a clip with synchronized sound, instead of renting a closed API. Distribution matters as much: once model files pass tens of gigabytes, whoever hosts them controls who downloads them, so torrent delivery spreads the burden across peers, echoing the point that the hosting layer is the real control point.

The widening cuts two ways: what a model does (forecast, see, hear) and how it reaches users. Licensing is the new dividing line, and TimesFM’s open code with restricted weights shows that “open” now needs a qualifier.

Read More: an earlier snapshot of the month’s trending open-source projects

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Open {Source} in {Late} {August:} {A} {Forecasting} {Model,}
    {Torrent-Shipped} {Weights,} and {Video} {With} {Sound}},
  date = {2026-09-08},
  url = {https://toknow.ai/posts/open-source-radar-timesfm-3-torrent-model-distribution/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Open Source in Late August: A Forecasting Model, Torrent-Shipped Weights, and Video With Sound.” https://toknow.ai/posts/open-source-radar-timesfm-3-torrent-model-distribution/.