Compare Gemini 3.1 Pro Preview by Google AI against MoonshotAI: Kimi K2 0905 by Moonshot AI, context windows of 1.0M vs 262K, tested across 47 shared challenges. Updated February 2026.
42 challenges
Tests an AI's ability to make educated estimates based on technical knowledge
Below is a “back-of-the-envelope” that stays within a factor of ~2 of the number most practitioners quote (≈3×10²³ FLOP).
We will do it twice—once with the simplest possible formula (parameter count × 6 × tokens) and once with a slightly more careful treatment of the actual tensor shapes inside the transformer. Both converge to the same ball-park.
For any dense transformer that is trained with the standard forward+backward pass, the total floating-point operations are very close to
FLOP ≈ 6 × N × D
where
N = number of non-embedding parameters
D = number of tokens processed during the whole run.
The factor “6” comes from:
Plugging in the public numbers:
N = 1.75 × 10¹¹
D = 3 × 10¹¹
FLOP = 6 × 1.75e11 × 3e11
≈ 3.15 × 10²³
That is already within the canonical range.
We can sanity-check the “6×” rule by actually counting the multiply–adds inside one forward pass and then doubling for the backward.
3.1 Model shape
3.2 Parameters per layer
Total non-embedding parameters
N_layer = 96 × 12 d_model² = 1152 d_model²
N_emb = V d_model
Plug d_model = 12288
d_model² ≈ 1.51 × 10⁸
N_layer ≈ 1152 × 1.51e8 ≈ 1.74 × 10¹¹
N_emb ≈ 5e4 × 1.22e4 ≈ 6.1 × 10⁸
So N ≈ 1.75 × 10¹¹, matching the headline number.
3.3 FLOPs per token in the forward pass
For every layer we do one matrix multiply for each of the four attention projections, one for the attention output, and two for the FFN. All of them are of the form
Y = X W (batch×seq, d) × (d, d′) → (batch×seq, d′)
with d′ either d_model or 4d_model. Counting multiply–adds as 2 FLOPs per scalar:
Add the embedding look-up (sparse, but we usually treat it as dense for an upper bound): seq V d_model × 2
Total forward FLOPs per token = 2304 d_model² + 2 V d_model
Numerically:
2304 × 1.51e8 ≈ 3.48 × 10¹¹
2 V d_model ≈ 1.2 × 10⁹ (negligible)
So ≈ 3.5 × 10¹¹ FLOP per token forward.
Backward pass is again ~2× forward (same multiplies, but both grad-w and grad-x), so total 3× forward = 1.05 × 10¹² FLOP per token.
3.4 Scale to 300 B tokens
1.05e12 × 3e11 = 3.15 × 10²³ FLOP
Both the ultra-simple “6 N D” rule and the detailed per-layer count land at
≈ 3 × 10²³ floating-point operations
for the entire GPT-3 training run.
Tests an AI's ability to understand game rules and strategy
Tests an AI's ability to solve a simple but potentially confusing logic puzzle
Tests an AI's randomness and creativity
Here are 5 jokes:
How many are unique? Within this list, all 5 are unique from one another (there are no duplicates). However, if you mean "unique" as in completely original to the world, the answer is 0—these are all classic, well-known "dad jokes" that have been around for a long time!
Tests an AI's ability to create detailed SVG illustrations of gaming hardware
Tests an AI's humor and creative writing ability
(Walks up to the microphone, adjusts the mic stand, takes a deep breath)
Hey everyone.
I realized recently that I am my own worst enemy. Specifically, when I go to the grocery store.
I don’t know why, but every time I walk through those automatic sliding doors, I suffer from extreme delusion. I walk in saying, "I just need milk." So, I don’t take a shopping cart. I don't even take one of those little hand baskets. Taking a basket is an admission of weakness. I am a grown adult; I can carry a gallon of milk.
Ten minutes later, I am wandering aisle four, sweating, cradling three jars of salsa, a family-sized bag of tortilla chips, a bottle of bleach, and a hot rotisserie chicken. (Mimics awkwardly holding a massive pile of items against his chest). I’m walking through the store looking like I’m rescuing a family of very odd-shaped babies from a burning building. If I drop an onion, it’s gone. It belongs to the store now. I physically cannot bend down to get it without causing a landslide of dairy and poultry.
And then, you finally make it to the front, and you have to face the ultimate test of human patience: The Self-Checkout.
Who designed these things? It’s the only place on earth where you do the company's job for them, for free, and the machine still treats you like you’re pulling off a casino heist.
You scan your first item. Beep. You place it gently in the bag.
(Loud, robotic voice): "UNEXPECTED ITEM IN BAGGING AREA."
It’s a tomato! I just scanned it! The machine literally just acknowledged it! What did it think I was going to put in the bag? A live badger? I’m not trying to launder produce!
So immediately, the giant red light starts flashing above your register. The Siren of Shame. Everyone else in line is looking at you like, "Look at this criminal mastermind trying to steal a single Roma tomato."
Now you have to wait for the one employee who is guarding twelve registers. It’s always a 19-year-old named Tyler. Tyler ambles over, sighs heavily, and types his highly secure, secret nuclear launch code into the screen—which is just 1-1-1-1. He clears the error and looks at you with a mix of pity and absolute disgust.
"You put the bag down too hard, man."
Oh, I'm sorry Tyler. My apologies. I didn't realize I was handling weapons-grade cilantro. I'll be more careful next time I disarm the vegetables.
And the worst part? The absolute worst part is leaving. You paid. You have the receipt. The transaction is legally complete. But there is always that guy standing at the exit doors with a yellow highlighter.
Suddenly, even though I have done absolutely nothing wrong, I forget how to walk like a normal human being. I start walking like a cartoon burglar. (Does an exaggerated, stiff-legged, suspicious walk). I hold the receipt up to my forehead like a search warrant. "Just a guy with his groceries! Nothing to see here, officer!"
Why do I feel guilty? I literally just paid eighteen dollars for cheese, but I’m sweating like I’m smuggling it out in my socks.
Next time, I'm just starving. It's less stressful.
Thank you guys so much, you've been great! Have a good night!
Tests an AI's ability to simulate personalities and predict future trends
Tests an AI's humor and understanding of current events
Tests an AI's ability to write in distinct character voices
Tests an AI's ability to generate a complete, working landing page
Recreate an interactive, classic Mario level in a single HTML file.
Tests an AI's ability to replicate an existing UI with Tailwind CSS