Please read the content of the review. I make it clear that GPT-5.2 Thinking is an incredible model, but I barely use it day-to-day, as Opus + Pro fit what I need much better, and speed lines up.
@mattshumer_
-
GPT-5.2 Thinking vs Gemini: Performance Comparison
By
–
GPT-5.2 Thinking is the overall winner, but Gemini is better at UI styling and other small things If you have a task you will do repeatedly, it's best to try them both for it
-
AI Models Code Generation Quality Less Critical as Capabilities Improve
By
–
1) that's not what I mean… it's just willing to work for longer without stopping, writing more of the system I want it to 2) code quality is going to matter less and less as models get better
-
Mixed Feelings About AI: Engagement Over Ignorance
By
–
To be clear, I love AI, but I do have mixed feelings about this all, personally. But ignoring it is not the answer, and will help no one.
-
Build Custom Solutions Instead of Generic Tools
By
–
There's zero need for them, and every use-case is different… it's way easier + better to build what you specifically need yourself
-
GPU Access and Sub-Agents Enhance AI Capabilities
By
–
Oh haha, maybe! Giving it GPU access + sub-agents makes it really capable
-
AI Model Harness Framework Shows Promise for Future Development
By
–
Obviously it’s not very capable (yet) but as models get better, a harness like this could be insanely powerful.
-
Autonomous AI Researcher Built with Gemini 3 Agent Swarm
By
–
Just built an autonomous AI researcher
— Matt Shumer (@mattshumer_) 21 novembre 2025
It uses a swarm of Gemini 3 agents with access to GPUs to allow it to run actual experiments and write a “paper”
Should I keep working on it / open-source it? pic.twitter.com/jmtzOnDQN8Just built an autonomous AI researcher It uses a swarm of Gemini 3 agents with access to GPUs to allow it to run actual experiments and write a “paper” Should I keep working on it / open-source it?
-
Model-Powered Prompt Harness Prototyping Strategy
By
–
Just calling my shot here… I’m pretty confident in this. Someone can prototype this today (maybe I will!) by having a model write a harness for a given prompt in Python, slot that into a @daytonaio sandbox or something similar, and then passing the prompt to the harness.
-
Self-Orchestrating Agents: The Next LLM Performance Breakthrough
By
–
Agents that natively self-orchestrate, managing their own context, tools, and sub-agents, are the next big unlock in LLM performance. Right now, a skilled engineer building an optimized harness, with thoughtful data flow, separation of concerns, sub-agent management, etc., can