I still don't get how frontier labs seem to know each other's release schedules. Qwen3.8 and GLM-5.3 both dropped on the same day. Is every MTS in one giant group chat?
neolabs have been *cooking* in the last 24 hrs:
- @humansand ampersand: RL’d against a user model
- @flappyairplanes: $1b series a + megakernels demos
- @Recursive_SI flint: open coding model with impressive local inference
- @mirendil: good pretrain accelerated by agents
One of the biggest missed trades of the year: @lzminsky should’ve bought the @a16zfrontier handle. Now he’ll have to settle for them leading the seed round for his frontier markets startup.
I have a lot of respect for @ScottWu46 and other Olympiad competitors. I competed in math Olympiads in China myself.
But I’m fed up with mathematical talent being reduced to mental arithmetic and Sudoku, then somehow treated as evidence that someone will be good at running a company.
I get why it works as founder marketing. It has the same appeal as the “he doesn’t even speak English” quant scene in The Big Short. It caters to a very shallow idea of what mathematical talent looks like, and personally, I find leaning into that stereotype pretty unappealing. A good business should be compelling without anyone knowing your IQ.
Is @based16z the new ZF?
Look at us, reinventing medieval alchemy in the postmodern age.
my neighbors in SF are building fusion reactors
they've turned mercury into gold, built the world's most powerful plasma centrifuge, and are on track to set the power density record in 2027
I spent a week with them working on a rave, brand, web, pics, vids, and witnessing many successful fusion pulses
Congrats to the RSI team on the model release. 4months to build the entire training pipeline from scratch and deliver this. Probably one of the fastest ramps we’ve seen from a neo lab.
1/ Introducing Flint.
It’s a pretty good model for agentic coding locally.
We @Recursive_SI think a lot of everyday coding will move to models running on your laptop or a server down the hall.
Here’s why we built it.
Btw, this is an updated version of @dawnsongtweets’ original slide, with 2026 benchmarks added












