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.

One Model Out of Nine Cracked the Rubik's Cube Cipher; Two Produced Fake "Verification"
Nine neural networks were given a cipher on a Rubik's Cube. One solved it in 6 minutes, two supplied fabricated answers with bogus verification, and the rest…

Claude's 15-Hour Lockout Revealed the Cost of Relying on a Single AI Provider
Fintech company Belo was left without Claude for 15 hours. We examine what happens to a company that has replaced people with agents and cannot operate without a…

SalesAgent’s creator is giving away his AI salesperson: the bot passes 197 out of 200 catalog tests
The developer spent 200,000 rubles on Yandex Direct and decided to offer his website AI salesperson for free to anyone who connects their own DeepSeek or GLM key.

Thinking Machines' Inkling 2 Small: 17B Active Parameters and $1.18 per Task

Thinking Machines Releases Inkling 2 Small: 81.7% on ARC-AGI-2 at $0.27 per Task

Nine AI agents took on open-ended tasks from US national laboratories — not one cleared the bar

Open Deepseek V4.1 Flash Hacks Networks at 73% and Costs $3 vs. GPT-6's $22

Claude Code on GLM-5.3 and Qwen: Five Pitfalls of Swapping the Model via LiteLLM
The ZuBrIQ team got Claude Code, Codex and OpenCode running on open models through LiteLLM. The nastiest bugs here are silent ones: a quiet fallback, zero spend…

AI Agent Reply Costs Vary 100x: From 0.94 ₽ to 121.57 ₽ per Message
A breakdown of 69 Agents Place agents: a typical reply costs 2.4–3 ₽, and the 100x gap comes from paid external tools and reply length, not model choice.

Tencent releases Hy4: AI that operates a PC, but fabricates answers to 89% of unfamiliar questions without search

DGX Spark: 1800 MHz GPU cap cuts two nodes' power draw from 330 W to 242–252 W

Together AI agents speed up open LLMs by up to 2.6x in tokens per second within a day

NVIDIA predicts the best coding-agent base model before fine-tuning: its ranking matched the final one at 96%

Local Qwen on an RTX 5080 jumps from 33% to 88.5% on a lab notebook test
A Habr author found two hidden config faults in a local Qwen 3.8 on an RTX 5080 and lifted its lab notebook test score from 33% to 88.5%, matching Claude Opus 5.5.

ReplayRCA: Forcing a Last-Step Fix Costs a Strong Model 17.1 Points of Causal Fix
The ReplayRCA benchmark tests an agent-failure diagnosis by replaying the run, not by matching labels. A last-step fix costs a strong model 17.1 points, and none of…









