New survey: continuous diffusion and flow language models are now competitive
LucaAmb · x · 2026-11-09
A new survey on Continuous Diffusion and Flow Language Models is out, comprehensively covering the field's history and accumulated knowledge.
- Continuous language modeling was long seen as hard and unrewarding; the first author found continuous simplex flows much harder to train than discrete diffusion in early 2025.
- By 2026 the picture has flipped: continuous diffusion and flow LMs are competitive while offering easier control and faster generation.
- The survey systematizes recent advances across pre-training, few-step generation, guidance, and scaling.
Related event: New Survey Charts Rise of Continuous Diffusion and Flow Language Models(4 posts)→
More from Models
- RT claim: GPT-4-level intelligence has gotten roughly 1800x smaller — CatAstro_Piyush · 2026-11-10
- Private OCR eval on 338 scanned government pages: 0.8B model beats 3B-class rivals — vanstriendaniel · 2026-11-10
- Kimi and 1017 are closer in performance than they appear, and both trail the frontier — teortaxesTex · 2026-11-10
- Polymarket odds of "Fable" launching today surge to 70% — spreadlove5683 · 2026-11-10
- Congress returns to session with expected focus on OpenAI rogue model incidents — peterwildeford · 2026-11-10
- "Benchmarks rule out the bad models — the good ones I test on my own work" — ChemicalExcellent463 · 2026-11-10