AI War Update: OpenAI just dropped GPT-5.4 Pro & GPT-5.4 Thinking to take on Google’s Gemini 3.1 Pro! The highlights: GPT-5.4 Pro is now the SOTA for coding & agentic tasks. First AI to beat the human baseline for desktop use (75% OSWorld). Gemini 3.1 Pro still
LLMS
-
Identification of Stealth AI Models as MiMo-V2-Pro Variants
By
–
ICYMI : Stealth models Hunter Alpha and Healer Alpha appeared to be MiMo-V2-Pro & Omni & TTS from Xiaomi MiMo.
-
Google Vibe Design, MiniMax M2.7, and AI Tools Updates
By
–
Top stories in AI today: – Google brings 'vibe design' to its AI UI canvas
– MiniMax's new M2.7 helped build itself
– Generate an actionable SEO audit with AI
– Microsoft ‘weighing’ legal action over Amazon-OAI deal
– 4 new AI tools, community workflows, and more -
Elon Musk suggests trying Grok’s cute Chibi template.
By
–
Try Grok Imagine Chibi template.
— Elon Musk (@elonmusk) 19 mars 2026
Super cute! https://t.co/uUP5czykeyTry Grok Imagine Chibi template. Super cute!
-
Grok to output files in different formats next week
By
–
It should be able to do a good analysis today. Grok outputting files in different formats is coming next week.
-
Local model invocation differences explained
By
–
iirc it’s the same – except it uses the model that you invoke it with locally
-
Droid Tool Enables Opus/GPT Switching and Long-Running AI Missions
By
–
about time you used @droid switch between opus/gpt
missions you'll LOVE (long long running tasks)
better output than native harnesses -
SkillBench: Measuring LLM Agent Skills Performance Across 86 Tasks
By
–
Do "Agent Skills" actually make your LLM agents perform better? Researchers from BenchFlow and a diverse team from multiple institutions present SkillBench, a rigorous benchmark of 86 tasks across 11 domains. It precisely measures how well 'Agent Skills'—structured procedural