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.

Opus plus DeepSeek Flash solved 67% of tasks versus 81% for Opus alone, and cost more
Ibragim Badertdinov tested an "Opus thinks, a cheap model codes" setup on 12 hard tasks. It scored worse (67% vs 81%) and cost more ($1.29 vs $1.20 per task).

DeepSeek Open-Sources Super-agent Prototype: p95 Task Time Drops from 29.5 to 15.7 Minutes
DeepSeek has released the code for Super-agent: a router talks to people while workers pick up GitHub tasks in parallel. In the company's test, p95 completion time…

NC State's AICG Framework Cuts MAE by 50.3% on Math Answer Grading
A RoBERTa model learns from students' skills as well as scores and down-weights AI-assisted answers. MAE fell by 50.3%, and QWK kappa rose to 0.927.

Strata Runs Qwen3.8-Flash-Next on a 12 GB RTX 5070: 94 Tokens per Second at Q2_0

Claude Enterprise Gets Smart Reports: Who Uses AI How, and What It Costs

243,000 texts for a 29M-parameter model: how the author picked data from Wikipedia and psychology books

WeFair-2B Query Rewriter Lifts CTR by 2.951% in WeChat Short-Video Search

OpenAI and Oracle to Build ChatGPT into Oracle Business Systems: Today's Other News
Oracle AI Workspace will connect corporate databases to ChatGPT. Also: Anthropic gave lawyers ready-made tools, Gemini 4 Carbon topped a ranking, and Microsoft made…

Lean Proofs for AI-Generated Code Break After Refactoring 13-27% of the Time
Researchers tested whether formal correctness proofs survive harmless code refactoring. Proofs from specialized AI models turned out to break more often than…

Study: A Single Random Failure Makes AI Agents Abandon a Working Tool 11.6% of the Time

Tencent releases Youtu-Parsing-Omni: one model parses documents, audio and video

AdaPRA: A Supervisor Agent Boosts AI Performance on Data Science Tasks by 24.5 Percentile Points

AOS-Bench tests AI agents: quality drops 13-23 points under parallel workloads

Cognition releases FrontierCode 2.0: the best score is just 45.5%
Cognition updated its coding-agent benchmark with larger tasks and stricter criteria. Claude Opus 5.5 scored 45.5%, while Kimi K3 managed only 17.7%, down from…

12 Pentesting Tools vs. Juice Shop: The Best AI Agent Found 5 Holes Out of 116
The author ran 6 AI agents, 4 DAST scanners and a Claude + Burp MCP combo against OWASP Juice Shop's 116 vulnerabilities. The hybrid won with 9/10; headline…









