this is really what happens when you leave your agents in a loop:
i gave the task of "port pokemon ruby to WASM and build a custom emulator to play it. do not use off the shelf emulators or port to C first. ROM binary to WASM"
then just let it cook.
full system access and the ability to use whatever tools and take as long as it wanted.
all in all it worked for ~4hrs, ran several hundreds tool calls, analyzed hundreds of errors, and made thousands of decisions along the way, with no help from me.
in the end i had a playable pokemon game in my browser, clean room reverse engineered from binary. 🤯
that is a loop, and not even the longest or most interesting i can imagine.
spawn 5.0 is out. we built all of it in a week.
million-particle storms. 1,000 lights. 100-player servers that multiply on their own. we're calling it "For Real" — that's what worlds look like now.
build anything with savi.
SOMEONE VIBE CODED A POKEDEX FOR REAL LIFE WITH CLAUDE CODE
point your phone at any animal, it identifies it on the spot and adds it to your personal index.
you walk around irl catching real creatures and filling out your dex like its gen 1 all over again
its called gotcha. every animal you spot gets logged with its own entry, so over time you build a living collection of everything youve actually encountered
it has:
> geo based rarity, a rabbit is common on a farm but legendary in a city
> profiles so you can show off your collection
> achievements for catching rare species
> battles or trading between players
its basically pokemon go all over again but with real animals
I've ran some tests on Gemma 4 12b on my RTX 3060 12gb
it seems to handle context pretty good and I have around 30tok/s almost all the way to 130k context
I asked GPT to test it out a bit for coding and comparing to Qwen 3.6 35B, and it seems quite close
I have tried the Q4_K_M from Unsloth, I want to see if the Q6 does any better quality wise
There is a Stone Age tribal war happening in Colombia right now
The Misak and Nasa people are fighting over a section of land. Yes this is real, and yes they are using traditional weapons. Video is from May 21st, 2026.
If you're waiting Gemma 4 12b through @ollama, its here:
gemma4:12b
gemma4:12b-it-q4_K_M
gemma4:12b-it-q8_0
gemma4:12b-it-bf16
gemma4:12b-mlxMLX
gemma4:12b-mlx-bf16MLX
gemma4:12b-mxfp8MLX
gemma4:12b-nvfp4MLX
https://t.co/8ptV7fVjNn
You'll need Ollama version 0.30.4-rc0
Running Gemma 4 12B locally at 28T/s. Facing issues with vision in LM Studio, but text performance is quite solid.
Asked it to create a landing page for a tech event.
Video is fast forwarded for the demo.
We’ve shipped a new release for the Antigravity IDE that significantly reduces cases of memory usage blowing up. We’ve been making a huge push to improve performance and reliability. The feedback is very helpful, we’ll get to all of it!
We’re launching Gemma 4 12B: Our unified, encoder-free model that brings powerful multimodal intelligence straight to your laptop 🚀
The model bridges the gap between our mobile E4B model and larger 26B MoE models, packaging frontier-class reasoning and native audio into a highly optimized footprint, all under a permissive Apache 2.0 license.
Here’s what makes it unique:
+ Encoder-Less Architecture: We removed the multimodal encoders. The vision and audio inputs flow directly into the LLM backbone.
+ Agentic Performance (16GB VRAM): Run complex, multi-step workflows locally, with performance nearing our 26B model.