Theo - t3․gg

Anthropic trained an evil version of Claude, and we need to talk about what that means. Thank you Clerk for sponsoring! Check them out at:...

Anthropic announced a 25% increase in Claude Code limits! That's good, right? It couldn't actually be a 17% decrease in usage...right? Than...

The fifth Model Context Protocol spec release from Anthropic last month changed the game and I want to use them way more now 👀 "stateless M...

Meta released their Muse Code powered by their latest model Muse Spark 1.2 and...it's interesting Thank you Greptile for sponsoring! Check...

Over 1,000 employees from OpenAI, Anthropic, and more, signed "Pacing the Frontier," a statement asking the US government to build the tools...

Jensen Huang's first-ever post on X shared an open-weights letter signed by practically every major AI company, except Anthropic - so we're...

Opus 5 is here Thank you Sent DM & Convex for sponsoring! Sent DM: https://soydev.link/sent Convex: https://soydev.link/convex Want to spo...

I once generated 10,000 lines of JavaScript just to organize 100 files without reading any of it, and it worked perfectly. That ratio is exa...

Claude Code's Co-Creator, Boris, just posted a HUGE breakdown on Twitter saying that the skills you spent years mastering when learning to c...

It’s time to breakdown the biggest difference between Fable and GPT 5.6-Sol, because while they are the biggest and best models right now, t...

GPT-5.6 Sol is genuinely better in Claude Code than Codex. I feel like I'm going insane. Thank you Coderabbit for sponsoring! Check them ou...

gpt-5.6 is here. I told you how I used it. Now I'll tell you how I feel about it. Thank you Posthog for sponsoring! Check them out at: http...

Codex is gone and has been renamed to ChatGPT Codex and ChatGPT Work under a single unified app, and I have thoughts. Thank you Blacksmith...

I got early access to GPT 5.6 before Fable even dropped, then burned somewhere between $180K and $240K of inference in six weeks to see what...

I don't think anyone expected Grok 4.5 to actually be good, much less outperform Opus 4.8 and rival GPT 5.5. But these are strange times and...

GLM 5.2 might be the best open-weight model ever released, but "running it yourself" means 400GB of VRAM, a $75,000 GPU box, and about $2,00...

Saying Fable has one-shot me would be a criminal understatemen; it's incredible, and honestly it might be better than I remembered. But if y...

As much as I love GPT 5.6, Ultra mode is a MESS because at the moment, one run can nuke your five-hour limit in 20 minutes because, while it...