LLM LAUNCHPAD. Beyond the chatbox. Built launch demos & new use-cases with foundation-model teams. Pushing models to the edge — or breaking them for what's next
Been building the demos behind model launches you've probably seen — quietly, with some great model teams. So many possibilities, poked and pushed and tested.
Sharing the work here now. 🧵 — Mike Luan, e01
⚡️ Step-3.7-Flash launch — pushed this flash-class model to the edge w/ @stepfun_ai.
Built Pointy (a second cursor that explains whatever you're looking at) + a set of parallel-research demos 👇
https://t.co/5AOqu1yEDi
Write up here:
https://t.co/iiGCZ7QYF6
With launch of Step 3.7 Flash by @StepFun_ai , we explored multiple new use-cases with "Flash Class" models.
The balance of speed, cost and quality gives life to new way to interact with the model.
This demo shows possibility to "Sample" the model in realtime, revealing layer of intelligence instead of a single "chat".
Write up coming soon.
Been building the demos behind model launches you've probably seen — quietly, with some great model teams. So many possibilities, poked and pushed and tested.
Sharing the work here now. 🧵 — Mike Luan, e01
consensus — ask once, a crowd of personas answers in parallel, weighted into one view. A model as a population, not a single reply. Built on GLM-5.1-Fast (+ other models) 👇
https://t.co/eqRW9K74E9
https://t.co/OBAQdRuBNv
lean & fast models + fast infra do make a difference.
in our latest experiment {consensus}, 50~100 persona (agents) are generated to answer one question.
turning "query" into "sampling", to potentially "extract" more from the model offerings.
https://t.co/eqRW9K74E9
repo:
https://t.co/D1QVwhF6Hu
A Lab note for Step 3.7 Flash launch.
--
When Flash models bring speed, cost, and intelligence into the “usable” range all at once, the way intelligence is supplied changes structurally.
--
https://t.co/WI9k8GwYiR
@StepFun_ai High speed, low cost and good quality + multi modal make latest "Flash" models a unique new class.
Harness are made for pro-models, Flash model turn all App into their harnesses.
https://t.co/VD9QGZOeXw
With launch of Step 3.7 Flash by @StepFun_ai , we explored multiple new use-cases with "Flash Class" models.
The balance of speed, cost and quality gives life to new way to interact with the model.
This demo shows possibility to "Sample" the model in realtime, revealing layer of intelligence instead of a single "chat".
Write up coming soon.
With launch of Step 3.7 Flash by @StepFun_ai , we explored multiple new use-cases with "Flash Class" models.
The balance of speed, cost and quality gives life to new way to interact with the model.
This demo shows possibility to "Sample" the model in realtime, revealing layer of intelligence instead of a single "chat".
Write up coming soon.
@jietang amazing performance, opens up new way to interact with the model.
from "chat" to "sampling" the latent of intelligence.
built during testing phase. - https://t.co/uK7X4cCrOF
lean & fast models + fast infra do make a difference.
in our latest experiment {consensus}, 50~100 persona (agents) are generated to answer one question.
turning "query" into "sampling", to potentially "extract" more from the model offerings.
https://t.co/eqRW9K74E9
repo:
https://t.co/D1QVwhF6Hu
We've added support for more model providers (openrouter | fireworks | gemini) to vcanvas and pushed source to github.
Grab the source here: https://t.co/wzQ1ZZRx99
Play with more providers here: https://t.co/huUtlGD4UG
Enjoy!
@witcheer Very fast and capable model, we compared it with Kimi 2.5 turbo while doing the demos and it is on-par, with even higher speed. A long waited one.
Collaborating with @Zai_org on early exploration around GLM-5V-Turbo (just launched), a strong, fast & multi-modal model.
Richer modality = more creative freedom.
We explored the interface side to enable it: a canvas...
makes visual interaction fluid — draw, collage, iterate, all in one space.
Pair that with 5V-Turbo's coding capability, you get sketch-to-running-code in seconds.
https://t.co/KjfrRHBmHi
Introducing GLM-5V-Turbo: Vision Coding Model
- Native Multimodal Coding: Natively understands multimodal inputs including images, videos, design drafts, and document layouts.
- Balanced Visual and Programming Capabilities: Achieves leading performance across core benchmarks for multimodal coding, tool use, and GUI Agents.
- Deep Adaptation for Claude Code and Claw Scenarios: Works in deep synergy with Agents like Claude Code and OpenClaw.
Try it now: https://t.co/WCqWT0qCQb
API: https://t.co/xDy1O6ZPcz
Coding Plan trial applications: https://t.co/qCM6cri0KK