DeepSeek
DeepSeek is a Chinese AI lab, backed by the High-Flyer hedge fund, that shook the industry with open reasoning models trained at a fraction of Western costs. Its R1 release in early 2025 wiped hundreds of billions off US tech stocks and made training efficiency a global priority. All our DeepSeek coverage is collected here.
NEWX: @0x0SojalSecSciforium pushes DeepSeek-V4.1-Flash on AMD to 1 129 tokens per second
Habr AIOne Chinese Comma Switched DeepSeek to Chinese in 94% of Responses
ZDNet AIAI Finds Unknown V8 Vulnerability for $4.24 in 13 Minutes, After Failing for $350 in February
Habr AIAI Agents Shipped 228 Game Releases Without Human Review; Tests Sent Back 25% of Tasks
Habr AITencent Hy4 flash scores 2.79 out of 4 vs. 2.81 for Haiku 5.5 and costs 18 times less per task
Habr AIFine-Tuning an NER Model Raised the Share of Fully Masked Deal Secrets from 39.5% to 91.8%
Habr AIAfter a Ban, the Author Launched an Agent Team in Kimi, Codex and OpenCode: Runs from $0.30
Habr AIQwen on two DGX Spark units runs 34–59% faster in tokens per second and outpaces GLM
Tencent Hy4 flash scores 2.79 out of 4 vs. 2.81 for Haiku 5.5 and costs 18 times less per task
Tencent's smaller Hy4 flash nearly matched Claude Haiku 5.5 in a blind test (2.79 vs. 2.81 out of 4), while one task costs ¥0.26 instead of ¥4.73.

Fine-Tuning an NER Model Raised the Share of Fully Masked Deal Secrets from 39.5% to 91.8%
The author fine-tuned the compact NER model GLiNER to mask only non-public discounts and payment schedules: fully hidden secrets rose to 91.84% from 39.53%, and…

After a Ban, the Author Launched an Agent Team in Kimi, Codex and OpenCode: Runs from $0.30
After Anthropic's ban wave, the supertick plugin author dropped his Claude Code dependency. His AI agent team now runs in Kimi, Codex and OpenCode; an OpenCode run…

How to find ChatGPT traffic in Metrica and GA4: up to 6,3% may be hiding in direct visits

Baidu Releases ERNIE 5.5 Flash Preview: 280B Parameters, 14B Active, 1M Context

MWS cut AI agent startup from 37.6 to 16.8 seconds: a harness teardown

ChatGPT Took 81.4% of Traffic Across Six AI Chatbots in Graphite Study

Developer Replaces Claude Code with Local Qwen and DeepSeek Models: Dev Cycle Gets 3–4x Longer
After Anthropic blocked his Claude Code account, a developer moved to open models on DGX Spark. It works, but tasks now take 3–4 times longer.

Nous Research Fine-Tunes Qwen3.8-27B: Propaganda in Answers Drops to 0.9% From 22%
Nous Research has released Arete-27B, a Qwen model that answers sharp questions about China with facts rather than the party line, without getting worse at coding…

Enthusiast Runs DeepSeek-V4-Flash on 64 GB of RAM With No GPU: Up to 15 Tokens per Second

Claude Fable 5.2 leads the MKA benchmark with 57%, but one task costs $31.97

llama-modchip runs a 284-billion-parameter DeepSeek in 64 GB without a GPU

DeepSeek Harness reaches 1.5 million users and 92T tokens a week

DeepSeek Shows Seven Prompts for Building Interactive Tools Without a Frontend
Instead of yet another chatbot: a video editor, a PDF viewer and a monitoring dashboard. DeepSeek has revealed the prompts Harness uses to generate working…

DeepSeek fixed 3 bugs in 1C in 104 minutes and reproduced two patches exactly
The agent received only the symptoms of errors in 1C:UNF, found their causes in large modules, and built extensions. It closed three bugs for RUB 155–427, but lost…









