Saw @DannyLimanseta and @sarmadsangi vibing 3D games and thought, what about the childhood classics?
I just built a 3D Connect 4 game in 3 prompts.
Not 3 hours. Not 3 days. 3 prompts and a few clicks in between...
Prompt typed:
3d connect 4 game
Can't see the board clearly
Increase the ball drop speed
See for yourself: https://t.co/8xGrRGcKAG
Dogfooding and cooking up @flavo_ai 👀
Want to build yours? Join the waitlist https://t.co/63npBDvm20
Inspired by @DannyLimanseta tiny skies! Here is what @flavo_ai wiped up for me in few prompts.
Not part of #vibejam but i thought i will still share some cool things @flavo_ai can build.
Play this game here:
https://t.co/5wL8KYplYz
This project management dashboard was single-shotted by our design agent. No templates. No Figma. No reference image.
You describe your app. We generate multiple designs. You pick the one that fits your taste. Then we build the whole thing.
We just opened our beta to a handful of new users. If you want to build apps that actually look like a designer touched them.
Before generating the actual app, users pick from themed visual mocks. No design work needed just choose a style and go.
This is the bento editorial theme, one of our favorite.
P.S. toying with a 3D floor plan app.
521M tokens routed across 26 models in 2025. Testing, iterating, breaking things, and rebuilding them better.
Still a work in progress, coming soon in January 2026!
You can now tell OpenRouter to heal JSON outputs from LLMs, across all providers. Results:
- Gemini 2 Flash: defect rate down by 80%
- Qwen3 235B: defect rate down by almost 100%
Got early access to Grok a few weeks back - thanks @xai team. Hitting around 200 TPS, which is impressively fast and economical, and it follows instructions really well.
Here's Grok generating 3 design mockups in 40 seconds.
I’ve also tried Grok Code as a Design agent for @flavo_Ai, an AI web app builder that I’m cooking with @sarmadsangi and @pengchongt.
You can see how fast it generated the Design mockups here. We went from 60-80 tps at best to 220-250 tps during our testing.
@claudeai@openrouter For planning, we have been toying between @GeminiApp Flash 2.5 for speed, and @OpenAI's new GPT5 for detailed planning.
Lately we have been toying with GPT5-mini to be a perfect balance between speed/latency and generating detailed plans.
Crazy how much we have spent in tokens testing our design agent last month! Our favourite is @claudeai Sonnet 4 and Opus 4.x so far.
Activity charts powered by @openrouter.
@levelsio@loaibassam Its a hard problem but at @flavo_ai we are attempting to change this, our design agent outputs web apps/pages that look distinct. We give you variations to select from before you go and build it out further.