#Open weights
10 articles tagged Open weights. Past halfway to the twelve-article threshold.
Timeline
-
Mistral Large 4 open weights — free to run 1T parameters, but API resale needs a license
-
Kimi K3.1 Pro: 0.7 points behind Claude Opus 5.5, with weights promised by November 4
-
Meta Muse Spark Open: 816 billion parameters you can download, under a license that isn't open source
-
What is model quantization: how a 1.5TB AI model shrinks to 4 bits and fits on one server
All articles
-
Tech · 5 min readMistral Large 4 open weights — free to run 1T parameters, but API resale needs a license
Release — weights on Hugging Face on October 29; about 1 trillion total and 52 billion active parameters, no download approval
-
Tech · 2 min readKimi K3.1 Pro: 0.7 points behind Claude Opus 5.5, with weights promised by November 4
Scale — ~2.8T total, ~104B active, 1M context, image and video input
-
Tech · 2 min readMeta Muse Spark Open: 816 billion parameters you can download, under a license that isn't open source
Scale — 816B total, ~42B active per token, 1M context, text/image/audio/video in
-
Tech · 2 min readWhat is model quantization: how a 1.5TB AI model shrinks to 4 bits and fits on one server
Idea — store each number in 8 or 4 bits instead of 16
-
World · 3 min readChina's AI export control draft — 23 of 40 technologies are AI, including model weights and agents
Scope — 40 technologies, 23 in AI; 14 prohibited and 26 restricted
-
Tech · 3 min readNvidia pays Poolside $6bn — the company selling the chips will now build the models
Size — $6bn license plus $1bn investment. Poolside was valued at $12bn pre-money
-
Tech · 3 min readNVIDIA's open voice model yields the floor in 0.48 seconds when you cut in
Replaces the ASR → LLM → TTS pipeline with one network, removing the latency between stages
-
Tech · 3 min readMeta put a 30B agent model on a laptop — Apache 2.0, no server required
Muse Glimmer 30B — distilled from Muse Spark 1.2, weights released under Apache 2.0
-
Tech · 4 min readOn-device AI explained — what changed to let big models run without a server
Memory is the binding constraint — cutting bits per parameter is what puts a 30B model on personal hardware
-
Tech · 2 min readWhat 'open-source AI' actually means — licences, weights, MoE
'Open-source model' is not one thing — openness runs from weights-only to full disclosure