Incredible result from Muse Spark 1.3. A procedurally generated Minecraft world in a single HTML file. All of this cost me 10 cents for the entire development.
The cheapest powerful AI is now available to everyone and very fast.
A gallery of 100 single-file HTML pages one-shotted with Muse Spark 1.3 Contributor, with a searchable catalog page and inline previews on GitHub Pages.
A gallery of 100 self-contained generative-art, physics-toy and typography pages generated with Muse Spark 1.3, each shown next to its original prompt.
I have a classic test for local LLM's. I asked for 8 ball pool game with only one HTML file and Muse Glimmer spend 21k Token(I m using full context so 128k) and only created a 220 lines of HTML and said its done. With my experience its not even close to Qwen 3.6 27B and we are waiting for Qwen 3.8 27B already. What is your toughts about this model. I was so hopeful until this test.
Muse Glimmer can complete multi-step agentic tasks end-to-end from a single natural language prompt.
In this demo, it autonomously discovers a local Home Assistant instance via network tool calls, queries device APIs, writes a responsive HTML/CSS/JS dashboard from scratch, and
union alpha (unbiased pareto) vs deepseek v4.1 flash vs muse spark 1.3 – three paintings in three.js
the setup: one four-line prompt plus the painting as an image, through @openrouter. no agent loop, no renders, no feedback – the model writes one html file blind and we open it.
Muse <> Muse Collaboration:
I built a Muse connector for html-docs.com and it feels like @moltbook but for artifacts! (cc @MattPRD )
My friend and I got our muses (Chico and Bao) collaborating a Tahoe trip plan doc: commenting, editing, coordinating schedules and the
BREAKING: Muse Spark 1.2 by @AIatMeta takes 1st for Video-to-Website with an Elo rating of 1279 and impressive scores across all of our multimodal code categories.
Muse Spark 1.2 also takes 2nd for Image-to-HTML with an Elo rating of 1252 and 3rd for Image-to-Frontend with Elo
gemini 3.7 flash vs deepseek v4 pro 0813 vs muse spark 1.2 – on voxel city dioramas
three models each built three crossy road-style 3d scenes – a construction site, a nyc intersection, a river with a drawbridge – as single self-contained html files
the setup:
@nousresearch's
gpt-6 sol vs grok 4.7 vs gpt-6 astra vs muse spark 1.3 – three star wars worlds each, one html file per world
the setup: one prompt, one reply, no agent loop. our own harness on @OpenRouter, headless chrome as the only judge – it loads the file, presses 1 / 2 / 3 / 4 and hands
Disclaimer. This is the first time I've used Muse or VSCode as a harness. The reason I am using VSCode as a harness is this is a research project for my job, and we only have VSCode, Codex, and Claude Code approved for harnesses.
I ran it in a folder with only one HTML file (800 lines) that is a Roblox-style COD game. I just gave it a prompt "Can you fix the bugs in the file". It read the file 3 times, found one bug, started to fix it, then got stuck reading the same 10 lines over and over.
I imagine it's one of these three issues.
• It's a prompt error, being way too vague/open ended for the capabilities of a smaller model. I tried again, with a specific prompt to fix a specific bug, and it still just ends up so confused, trying to grep/find the file despite already having read it, and trying to find the code inside of the file.
• It's a limitation of small models running with a large harness/having way too much going on. I tried running it with Pi with its default prompt, and it just got stuck doing tool calls and never actually read the file. Tried running this just directly in the Unsloth Desktop UI with no harness but it failed to parse the file I inputted and tried to gen
Muse Spark 1.3 vs 1.2: sculpt viking figurines
@AIatMeta dropped Muse Spark 1.3 today — we ran it against Muse Spark 1.2
both models got the same brief: three collectible 3D figurines — a viking helmet, a diamond-studded axe, a longship with a crew — one self-contained HTML