I gave Kimi K2.6 and K2.7 Code the exact same prompt to animate a letter burning to ash. Pure HTML canvas, zero libraries.

Fable 5 Max for you...

Whoa

Prompt: Create a single HTML file with a canvas animation of a handwritten letter burning. Show an aged, slightly yellowed sheet of paper with visible handwritten cursive text (procedurally drawn lines are fine) resting on a dark wooden desk. After 2 seconds, a flame ignites at the bottom-right corner and spreads organically across the page — the burn front should advance with an irregular, noisy edge, never a straight line. Just ahead of the flames, the paper should darken and brown (scorching), then char black, then disappear entirely, revealing the desk beneath. Render the fire with layered particles: a bright white-yellow core, orange mid-flame, and translucent red tips that flicker and lick upward. Glowing embers should detach from the burn edge and drift upward on turbulent air currents, fading from orange to gray. Add wisps of semi-transparent smoke rising and dispersing above the flames, and a warm flickering light that the fire casts onto the surrounding desk. The entire page should be consumed in roughly 15 seconds, leaving only a few glowing ash fragments that slowly dim. 60fps, no external libraries.

K2.6 looks better?

Honestly both are solid, but I'm really into the ignition randomness of K2.7 Code.

I like the burning effect more of 2.6. 2.7 seems to add random sparkles with no realistic positioning imo

Tried with Gemini 3👀

This is awesome!

qwen-3.6-27B comparison .. this is a 37x smaller model :)

fire ignition from k2.7 code is dope

K2.6 is the most accurate and realistic. It simulates the actual way such a paper would burn. @jakeharrisdev used fable 5 Max which is trash but it got one factor right. It left ashes which K2.6 omitted. K2.6 gets the combustion right. Fable 5 max gets the after effects right Speaking from experience during my Prometheus phase.

We should tar and feather whoever started this one-shot obsession for showcasing LLMs. And we should do the same to any person who shows n=1 when doing it themselves.

Love this one! Stevibe super Hero!

K2.6 looks more realistic

K 2.7 is good. Probably not the best example for it, cause here, 2.6 looks better/more realistic

I've burnt paper , 2.6 is more realistic

The THINKING TRACES man....!!! K2.7 Code is very similar to Nex-N2 Pro!! - It uses "We" instead of "I" - Uses "Need" extensively - Uses "maybe" extensively - Concise, caveman style Very interesting.

2.6 is better

2.6 seems a bit more realistic, I also like the fact that it left some after effect (shadow) after burning

Zero libraries and it pulls off fire physics like that 👀 Wild how much K2.7 Code improved in just one version.

The initial reactions of Kimi K2.7 are really good. Can't wait to try it. @opencode can you give me free access to it? 😁

Real letter burners know Kimi K2.6 was more realistic.

right one feel bit realistic...

actually the old one seems better 0_0

drop the results i want to see the lag

I hope opencode gets this on the go plan soon

There's a clear improvement.

Both are pretty bad haha, k2.7 is much less so though. Very interesting problem! How do claude models fare on this benchmark?

what provider are you using?

OpenRouter -> MoonshotAI

Kimi 2.6 did it better than 2.7

Wow very different

More realistic 2.6

GLM-5.2

liked 2.6 better in terms of looks did you track how much time each took? did you have to correct or was it a one shot?

watched identical canvas burn effects diverge once mobile forced a 2k particle budget

@ivanfioravanti Could you count token usage for both?

@ivanfioravanti I didn't count it this time. I'll have it next time.

2.6이 더 깔롱한데?

kim 2.7 is really good. Test it too and was amazed.

Wait, wait, wait, wait, and wait! K2.7 Code? WTH?!