This Week in DeepSeek

A weekly digest of the last four weeks: what the community is discussing, what papers came out, what the Chinese community is talking about.

How this is made. No newsroom. We aggregate public content from Hacker News, arXiv and CSDN, group it by week and link the originals. Refreshed weekly, not real-time.

We do not judge what matters — items are ordered by heat (HN points) and recency.

Weeks covered

Items on page

Refresh

weekly

Editorial

none

2026-11-02 ~ 2026-11-08 (this week)

16 items

Community (Hacker News)

Launch HN: Momentic (YC W24) – Agents that catch bugs by using your app	166
Ten harness patterns that you can apply to a real repo	12
DeepSeek Flash v4.1 Explained Through Harry Potter	3
DeepSeek Harness [video]	2
DeepSeek alone now accounts for 53% of all token volume on Openrouter	2
Over 500k words in a single DeepSeek thread	2
Evaluating DeepSeek v4.1 Flash-1017 for loss of control and harmful manipulation	2
DeepSeek also can easily read and write ML Weights	2

Research (arXiv)

Trace2Eval: Contextualized Evaluation of Coding Agents from Real-World Traces

SafeStart: Mitigating the Overlooked Safety Cost of Active Expert Reduction in MoE LLM Serving

SBVUL: Statement-Based Vulnerability Detection

vExpert: Virtualizing Expert Storage for Adaptive Load Balancing in Distributed MoE Inference

How Well Can Large Language Models Impersonate Individual Authors? A Comparative Evaluation

Iterated creation, grading, and revision of data science projects by language models

LeanSparse: Memory-Efficient Serving for Native Sparse-Attention LLMs

GC-MoE: Overlapping Expert-Parallel Token Dispatch with FP8/FP4 Grouped GEMM via CUDA Green Contexts

2026-10-26 ~ 2026-11-01 (last week)

16 items

Community (Hacker News)

There are now 5 inference providers for abliterated DeepSeek	100
DeepSeek open-sources dsh-superagent, a coding agent largely built by agents	11
DeepSeek put the architecture for their always-on autonomous agent on GitHub	7
DeepSeek-v4.1-Flash: Pushing the Limits of KV Cache Compression	6
DeepSeek reportedly raises $15B days after US restrictions	5
DeepSeek launches advanced coding agent to rival Claude Code	4
Show HN: Fastbrowse, Browser agent that picks clicks with a System One model	4
DeepSeek Harness Superagent Prototype	3

Research (arXiv)

Multi-Agent Collusion and the J-Space

Recursive Self-Improvement of Exoskeleton Assistance Policies through Human-in-the-Loop Feedback

\textsc{ChatDify}: Automating End-to-End Dify Workflow Synthesis from Natural Language via Multi-Agent Collaboration

Do GEO Gains Survive the Live Web? Evaluating In-the-Wild Visibility Gains of Generative Engine Optimization via LIVE-GEO

CompSkill: Compiling Execution Experience for Skill Retrieval

Decision-Centric Research: A Calibrated Decision Model and Benchmark for Scientific Agents

Precedent Labels and Facts Shift LLM Approval Decisions in Opposite Directions

Enhancing Language Model Reasoning with Structured Multi-Level Modeling

2026-10-19 ~ 2026-10-25

16 items

Community (Hacker News)

DeepInfer: High-performance inference framework for DeepSeek models	2355
DeepSeek banned from U.S. federal civilian agencies under new DHS order	584
DeepSeek V4.1 Flash 1017 (Max)	191
The US orders every civilian agency to rip out DeepSeek, even tools built on it	66
Alibaba hikes DeepSeek V4 Flash 0731 output prices by 230%	26
Nexus-CLI – Pure Go Bridge to DeepSeek R1/V3 with Claude Code and Ollama Support	15
DHS Orders DeepSeek Technology Removed from Federal Civilian IT Systems	9
DeepSeek just raised $15 billion to take on OpenAI and Anthropic	8

Research (arXiv)

AgentProofBench: Benchmarking Reusable Verifiers for Web Agents

Speculative Decoding for Mixture-of-Experts Models: Expert Bytes, Kernel Bandwidth and the Best Draft Length

Improving the Chain-of-Thought Compressibility of Pruned Reasoning Models via Long--Short Contrastive Pruning

From Document Layout to Causal Topology: An End-to-End Architecture for Temporal Causal Reasoning

ProtAgent: Post-Training Compact Protein Tool Agents with Interaction-Level Credit Assignment

Multi-Agent Retrieval-Augmented Causality Assessment (MARCA) of Individual Case Safety Reports in Pharmacovigilance

AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science

HarnessMM: Recursive Self-Improvement for Harness Token Efficiency

2026-10-12 ~ 2026-10-18

16 items

Community (Hacker News)

DeepSeek-v4.1-Flash-1017	959
DeepSeek search and fetch – $0.75 per 1000 successful requests	607
DeepSeek: DeepSeek v4.1 Flash 1017	520
Show HN: Just gave Claude Code the website and ad budget to see what it would do	11
DeepSeek-v4.1-Flash-1017 is here	9
DeepSeek Outage	6
DeepSeek-v4.1-Flash-1017: Wish-Driven Programming Made Real	4
DeepSeek v4.1 Flash is now free in OpenCode	4