DeepSeek V4.1 Flash Confirmed to Center on YOCO Architecture

On Sept 10–11, commentators focused on the architecture of DeepSeek's V4.1 Flash. DeepSeek researcher Dong Li confirmed the model's soul is YOCO (You Only Cache Once), inheriting its order-of-magnitude advantages in prefill and KV cache; close readings of the paper also surfaced KV cache reuse across checkpoints, and the related techniques are fully open-sourced. TL;DR: V4.1 Flash's core gains come from YOCO's prefill/KV-cache optimizations, and DeepSeek is credited with pushing architecture-level innovation into the open-source community.

Confirmed

• Dong Li (@donglixp) said V4.1 Flash's architecture centers on YOCO and that architectural innovation has entered its second half.
• Astra Pro's paper deep-dive found KV cache reuse between checkpoints, with one commenter calling DeepSeek a "weird model factory."
• The DeepSeek paper cites the 2024 YoCo paper as the main inspiration for its Causal-Encoder Decoder (CED) transformer module.
• The new architecture's techniques are publicly open-sourced.

Unconfirmed

• Chinese KOL MaxForAI's (unofficial) reading of the paper's diagrams suggests at least three structural changes, including a back-and-forth encoder-decoder layout and CSA2 cross-layer reuse; the real story is changes to the Transformer itself, not parameters or benchmarks.
• Developer aHapBean argues the "Causal Encoder-Decoder" label overstates the novelty: V4.1 may still be a decoder-only LLM, better described via the YoCo framing.

Why it matters

• @evijit notes DeepSeek keeps innovating at the architecture level, not just scale, and open-sourcing it raises the floor for the whole open-source ecosystem.
• If order-of-magnitude KV cache optimization and cross-checkpoint reuse prove out, they could meaningfully cut training and deployment costs and directly improve inference economics.

2026-09-10 ~ 2026-09-11 · 7 related posts

Full story(18 episodes)→

• Episode 1: DeepSeek V4.1 Flash Opens Limited Internal Beta with New Architecture and Native Multimodality(2026-09-08, 18 posts)
• Episode 2: DeepSeek V4.1 Flash Tested: Blazing 350 Tokens/s but Still Experimental(2026-09-08, 2 posts)
• Episode 3: DeepSeek cuts V4-Flash API prices with new peak/off-peak billing from Sept 10(2026-09-08, 6 posts)
• Episode 4: DeepSeek V4.1 Flash Nears Frontier Models at a Fraction of the Cost, Independent Tests Show(2026-09-09, 15 posts)
• Episode 5: DeepSeek Releases Open-Source V4.1 Flash, Beating Its Own Flagship at a Fraction of the Cost(2026-09-09, 57 posts)
• Episode 6: DeepSeek V4.1 Flash Leaks: 552B Asymmetric MoE Claimed to Rival GPT-5.6 Sol(2026-09-10, 21 posts)
• Episode 7: DeepSeek V4.1 Flash Benchmark Results Leak, Sparking Community Discussion(2026-09-10, 2 posts)
• Episode 8: DeepSeek Shrinks Per-Token KV Cache 54x in Nine Months to Sub-KB Levels(2026-09-10, 6 posts)
• Episode 9: Leaked DeepSeek-V4.1-Flash Touted as New Open-Source King(2026-09-10, 3 posts)
• Episode 10: Bug Hunt Bench Ranks Frontier Coding Models on 105 Real Bugs(2026-09-10, 13 posts)
• Episode 11: DeepSeek V4.1 Flash Confirmed to Center on YOCO Architecture(2026-09-10, 7 posts)
• Episode 12: Leaked DeepSeek V4.1 Flash Tech Report Reveals KV Cache-Obsessed Flagship(2026-09-11, 44 posts)
• Episode 13: DeepSeek V4.1 Flash tops Vals open-source index at $0.30 per run(2026-09-11, 2 posts)
• Episode 14: DeepSeek launches V4.1-Flash: 552B backbone, 1M-token context(2026-09-11, 5 posts)
• Episode 15: DeepSeek V4.1 Flash scores 40 in AA index despite rising hallucinations(2026-09-11, 2 posts)
• Episode 16: Inside KV-Cache Sharing: How DeepSeek CED and GLM 5.2 Differ(2026-09-11, 4 posts)
• Episode 17: DeepSeek 4.1 Flash Shows Big Gains in Kernel Engineering and Cost(2026-09-11, 2 posts)
• Episode 18: SGLang delivers Day 0 support for DeepSeek V4.1 Flash, hitting 873 tok/s(2026-09-12, 2 posts)

• DeepSeek returns to encoder-decoder with new open architecture changes — evijit · 2026-09-10
• [source] DeepSeek's V4.1 Flash reworks the Transformer again: asymmetric encoder-decoder and CSA2 — evijit · 2026-09-10
• DeepSeek cites 2024 YoCo paper as inspiration behind its CED transformer blocks — jm_alexia · 2026-09-10
• [source] DeepSeek's V4.1 paper reveals KV cache reuse across checkpoints — 'a very strange model factory' — teortaxesTex · 2026-09-10
• DeepSeek V4.1 'causal encoder-decoder' label debated: likely a YOCO-style KV-reuse design — donglixp · 2026-09-10
• DeepSeek V4.1 Flash reportedly adopts YOCO architecture for order-of-magnitude KV savings — donglixp · 2026-09-11

1 near-duplicate retellings: donglixp