Head to head: Kimi K3 vs xAI: Grok 4.5
This was a thin matchup on paper, but Kimi K3 still comes out ahead on the numbers. The catch: the only judged task showed enough order sensitivity that this result reads as a lean, not a rout.
By RuntimeWire · Published
Kimi K3 takes this head-to-head on the aggregate, 8.6 to 4.2, and the statistical read backs that up with a 75% confidence lean. That is enough to put it in front, but not enough to pretend this was a demolition. With only one scored task and no ties, the margin looks cleaner than the evidence really is.
On the actual prompt — "A forgotten vending machine in..." — Kimi K3 got the official task win because it more clearly assembled the requested scene: moonlit forest, readable vending machine, glowing fireflies, raccoons, and visible animated activity. The key advantage was legibility. It made the environment and motion readable instead of burying them in darkness.
But the order-swapped judge pass matters here, and it muddies any triumphalism. In that pass, Grok 4.5 was preferred for stronger composition and richer lighting, while Kimi K3 was described as too dark and sparse. That kind of reversal is a warning sign: the matchup did not expose a stable, overwhelming quality gap so much as a narrow edge that depended on which rendering better surfaced the prompt details.
So the editorial verdict is straightforward: Kimi K3 was the better performer in this test, mostly because it was more successful at making the scene readable and complete. Grok 4.5 did not put up enough on the board to challenge the aggregate result, but the evidence is still thin and a bit noisy.
Final call: Kimi K3 wins — deservedly, but only as a lean.
How they were tested
We ran 1 fresh anim task, generated on the fly for this matchup so neither model could prepare in advance, and had gpt-5.4 score each one. Kimi K3 scored 8.6 to xAI: Grok 4.5's 4.2.
1. A forgotten vending machine in
A forgotten vending machine in a moonlit forest dispensing glowing fireflies as curious raccoons gather around it.
Winner: Kimi K3 — Model A clearly delivers the moonlit forest setting, readable vending machine, glowing fireflies, and raccoons with appealing composition and visible animated activity. Model B includes the core objects but is so underlit and sparse that the forest, fireflies, and motion are barely legible, reducing both thematic clarity and polish. (Order-swapped judge pass: Model B clearly delivers the moonlit forest setting, readable vending machine, glowing fireflies, and raccoons with stronger composition and richer lighting. Model A matches parts of the prompt but is extremely dark and sparse, making the scene and motion harder to read despite having raccoon-like figures around a glowing machine.)
See every prompt and the full side-by-side outputs in the interactive Head-to-Head.