Pinned Tweet
Gemini fans, look away🤓
Same prompt: "3D human anatomy atlas".
Gemini 3.8 Flash: floating dots.
GPT-6 Astra: a full human body. Muscles, bones, organs. Click any part and it tells you what it is.
Med students would love this. It runs in a browser tab.
Prompt in the first reply 👇
Gemini or Astra? Reply 1 or 2
Aimore retweeted
Claude just turned one chat into a workspace🤯
Sheets builds spreadsheets
Whiteboards turns ideas into plans.
Data builds dashboards from connected sources.
Partner apps open inside Claude. Meetings is already in beta.
AI isn’t just answering questions anymore.
Claude just turned one chat into a workspace🤯
Sheets builds spreadsheets
Whiteboards turns ideas into plans.
Data builds dashboards from connected sources.
Partner apps open inside Claude. Meetings is already in beta.
AI isn’t just answering questions anymore.
Official Anthropic demos and details:
Sheets, Whiteboards, Data:
claude.com/resources/article…
20
Aimore retweeted
Claude may have helped crack a 167 year old math problem👴
Anthropic mathematician Levent Alpöge claims a proof of the Riemann Hypothesis, plus its generalized form.
He released 14 Lean archive parts.
But no independent verification yet.
This could be historic. Or wrong
🤖 Made with AI
Claude may have helped crack a 167 year old math problem👴
Anthropic mathematician Levent Alpöge claims a proof of the Riemann Hypothesis, plus its generalized form.
He released 14 Lean archive parts.
But no independent verification yet.
This could be historic. Or wrong
🤖 Made with AI
Aimore retweeted
Claude Fable 5.2 helped build a swimmable underwater world in a 96 KB HTML file🌊🌊
No images. No 3D models. No game engine.
Just WebGPU: kelp forests, manta rays, a shipwreck and underwater caves.
An entire explorable ocean, smaller than many profile photos
Claude Fable 5.2 helped build a swimmable underwater world in a 96 KB HTML file🌊🌊
No images. No 3D models. No game engine.
Just WebGPU: kelp forests, manta rays, a shipwreck and underwater caves.
An entire explorable ocean, smaller than many profile photos
Open-source code:
github.com/Noniv/abyssal-gar…
35
Aimore retweeted
3 AI models tried to build a rocket launch simulation🚀
Same challenge. Wildly different costs.
--> Claude Opus 5.5: $49.07
--> GPT-6 Astra: $16.55
--> GPT-6.1 Sol: $2.05
Opus produced the creator's favorite result.
But Sol delivered a surprisingly convincing scene for nearly 24x less.
The real AI race isn't just about quality anymore. It's about what that quality costs
3 AI models tried to build a rocket launch simulation🚀
Same challenge. Wildly different costs.
--> Claude Opus 5.5: $49.07
--> GPT-6 Astra: $16.55
--> GPT-6.1 Sol: $2.05
Opus produced the creator's favorite result.
But Sol delivered a surprisingly convincing scene for nearly 24x less.
The real AI race isn't just about quality anymore. It's about what that quality costs
Aimore retweeted
Same 3D challenge. Two very different results🤨
DeepSeek V4.1 Flash vs GPT-6.1 Sol.
--> DeepSeek: $0.96
--> GPT: $1.29
But here's the catch.
DeepSeek took over 8 hours.
GPT finished in about 21 minutes.
One looks more detailed. The other is dramatically faster.
Which result would you actually choose?
Same 3D challenge. Two very different results🤨
DeepSeek V4.1 Flash vs GPT-6.1 Sol.
--> DeepSeek: $0.96
--> GPT: $1.29
But here's the catch.
DeepSeek took over 8 hours.
GPT finished in about 21 minutes.
One looks more detailed. The other is dramatically faster.
Which result would you actually choose?
Aimore retweeted
One developer. Claude Opus 5.5. 25,000 players in 2 weeks✈️
That's the creator's claim for VIPER 11, a browser fighter-jet game with F-16s and ranked 1v1 dogfights.
He used multiple AI agents on separate code branches, then reviewed their work.
No install. Just fly.
Aimore retweeted
48 AI agents scanned 1,200 km² of Maya jungle using NASA lidar🏝️
They flagged a plaza, a pyramid-like structure and a courtyard. Mexico’s INAH had no site registered there.
The twist: archaeologists had already mapped it in 2025.
AI found the shapes. It missed the paper
48 AI agents scanned 1,200 km² of Maya jungle using NASA lidar🏝️
They flagged a plaza, a pyramid-like structure and a courtyard. Mexico’s INAH had no site registered there.
The twist: archaeologists had already mapped it in 2025.
AI found the shapes. It missed the paper
Original experiment, correction and visuals:
pietromiotti.com/latest/maya…
56
Aimore retweeted
AI agents are being tested on work that takes experts 3-8 hours🤖
Google and Artificial Analysis built MKA, a benchmark with 200 complex tasks across 83 occupations.
Not trivia. Real workflows involving documents, diagrams, spreadsheets and reports.
Gemini 4 Carbon ranked #2 in quality, while costing 6x less per task on average than the top-scoring model.
My take: the next AI race isn't just about benchmark scores. It's about how much useful work an agent can deliver for the money.
Important: these are benchmark results, not proof that AI can replace an expert.
What matters more to you: maximum accuracy or lower cost per task?





