A tokenization-shared open model ladder is sorely missing, and all roads lead to Qwen
bclavie · x · 2026-10-22
Researcher bclavie argues there's huge value in a neolab releasing a good 2-to-18B dense / 15-to-40B MoE ladder with a tokenizer aligned to GLM's, DeepSeek's or Qwen's. A tokenization-shared model ladder for experiments is sorely missed right now—which is why 'all roads lead to Qwen' for researchers running experiments.
More from Models
- DeepSeek-V4.1-Flash spotted running 5 parallel research subagents at near-zero cost — teortaxesTex · 2026-10-22
- xAI adding 200,000 GB300s a month, 10GW of compute planned next year — gaganghotra_ · 2026-10-22
- Anthropic most complete, OpenAI most accurate: 14 models compared on medical benchmarks — ArtificialAnlys · 2026-10-22
- BitKindle: AI bootstraps a full C compiler from a 3,584-byte machine-code seed — dotey · 2026-10-22
- Chonk DOOM: a DOOM-like shooter built for Mistral Large 4 to beat on its own — GuillaumeLample · 2026-10-22
- DeepSeek V4.1 Pro rumored to train on 50T+ pretraining tokens, matching Microsoft's scale — teortaxesTex · 2026-10-22