한국어EN
Model Architecture & Training MethodsKeller Jordan

Keller Jordan releases the Muon optimizer

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Muon optimizer released

Why it matters

Set new training-speed records over AdamW using Newton-Schulz orthogonalization

Source — Primary sources first; secondary reports are labelled “Press”

Officialkellerjordan.github.iohttps://kellerjordan.github.io/posts/muon/

Added in Week 35, 2026 · See that week

On the map — ⑪ Model Architecture line

  1. Previous stopDeepSeek-V2
  2. This recordMuon optimizer
  3. Next stopByte Latent Transformer
See it on the map

Same week

Meta releases Llama 3.3 70BOfficialOpenAI launches ChatGPT Pro at $200 a month, full o1 and o1 pro modeOfficialBIS expands the FDPR, adds HBM controls and 140 Entity List entriesGovernmentNous Research releases DeMo paper and code; 15B model trained with DisTrOOfficial

Nearby on the same axis

Byte Latent TransformerPaperDeepSeek-V3 Technical ReportPaperNSA — Native Sparse AttentionPaperNaver releases a new HyperCLOVA X modelOfficial