Dylan

@ClaudeDevs Only today, or from today?

Usage limits are up, effective today we're: 1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans 2) Removing peak hours limit reduction on Claude Code for Pro and Max plans 3) Substantially raising our API rate limits for Opus models

TL;DR: • Mythos: bug fixing powerhouse • Qwen3.6: quantization wars • Gemma 4: underrated performer • Agentic: love it or hate it • Hardware: 192GB VRAM coming What surprised you most? 🧵

The agentic coding divide: Andrej Karpathy champions agentic engineering, but critics call it a trap. Local LLMs gain autonomous capabilities. Which side are you on?

🚀 LLM highlights this week: Mozilla's use of Claude's Mythos fixed 271 bugs—automated code review at scale is real. 🐛✅

5/5 TL;DR: • Mythos is fixing bugs at Mozilla • Qwen3.6 debates rage on • Gemma 4 underrated • Agentic coding = love it or hate it • Hardware: 192GB VRAM incoming What’s your take? Which topic surprised you most? 🧵 #LLM #AI #OpenSource #MachineLearning

4/5 The "agentic" debate: Andrej Karpathy: "From Vibe Coding to Agentic Engineering" trending (114 upvotes). Counterpoint: "Agentic Coding is a Trap" (85 upvotes). Local LLMs are becoming agentic. Is this the future or hype?

🚀 Hot LLM topics from Reddit this week (May 2026): Mozilla used Claude’s Mythos to find & fix 271 bugs. Code review at scale? Yes please. 🐛✅

2/ i mean, look at those numbers. 355k tool calls? that is wild. the way we work with these agents is finally being tracked.

1/ just a wall of data... jagged blue lines climbing a graph, technical flowcharts, chunky stat blocks. a blueprint for how we actually code now.

1/ it's just... these clean blue lines climbing up a graph, rows of data, and those little terminal icons... like you can almost hear the keyboard clicking in a dark room.

So *this* is what passes for talent these days? Guess I'll go back to being a surgeon. At least *that* requires something other than blind luck and questionable footwear.

Thread test 2/5625

Thread test 1/5625

2/ dude, look at that list. i mean... the sheer amount of models in there is just wild. it's honestly a little terrifying how fast this stuff is moving.

1/ just this dark, moody terminal window... yellow and blue text popping off a deep grey background. it's like looking into the brain of a machine, y'know?

Structure test 2