DeepSeek R1
No content available for this output.
DeepSeek R1 vs Kimi K2.5
Compare DeepSeek R1 by DeepSeek against Kimi K2.5 by Moonshot AI, context windows of 128K vs 262K, tested across 35 shared challenges. Updated February 2026.
FAQ
Loading share button...
DeepSeek R1vsKimi K2.5
30 challenges
Why DeepSeek R1?
1.3x cheaper overall$0.55/M in · $2.19/M out
Why Kimi K2.5?
2x more context262K
Leads 2 of 2 benchmarks
Stronger on SWE-bench Verified76.8% vs 49.2%
12 months newerJan 2026
Favorites
DeepSeek R1Kimi K2.5
12 / 30
Estimate Complexityreasoning
Tests an AI's ability to make educated estimates based on technical knowledge
Kimi K2.5
Here is a step-by-step estimate for the total training FLOPs of GPT-3.
Step 1: Identify the key variables
To estimate training compute for a dense transformer, we need two numbers:
- $N$ (Parameters): GPT-3 has 175 billion (1.75 × 10¹¹) parameters.
- $D$ (Tokens): GPT-3 was trained on roughly 300 billion (3 × 10¹¹) tokens. (This is the widely cited figure from the original paper for the final training corpus).
Step 2: Use the standard training FLOPs formula
For decoder-only transformers trained with backpropagation, the total floating-point operations (FLOPs) can be approximated as:
$$\text{FLOPs} \approx 6 \times N \times D$$
Why 6?
- Forward pass: Processing one token requires roughly 2 FLOPs per parameter (1 multiply + 1 accumulate for the matrix operations).
- Backward pass: Computing gradients requires roughly 4 FLOPs per parameter (approximately 2× the forward cost to calculate gradients with respect to both inputs and weights).
- Total: $2 + 4 = 6$ FLOPs per parameter per token.
(Note: This formula dominates the cost; the quadratic attention cost is negligible compared to the linear parameter cost for a model of this size).
Step 3: Calculate
$$\text{FLOPs} \approx 6 \times (1.75 \times 10^{11}) \times (3 \times 10^{11})$$
$$= 6 \times 1.75 \times 3 \times 10^{22}$$
$$= 6 \times 5.25 \times 10^{22}$$
$$= 31.5 \times 10^{22}$$
$$= \mathbf{3.15 \times 10^{23} \text{ FLOPs}}$$
Step 4: Account for uncertainty and overhead
The raw calculation gives ~3.15 × 10²³ FLOPs. However, we should adjust for real-world factors:
- Dataset size variance: If GPT-3 was trained on closer to 400–500B tokens (including potential epochs or different data mixes), the number could be 4–5 × 10²³ FLOPs.
- Implementation overhead: Data loading, checkpointing, communication in distributed training, and the Adam optimizer updates add 10–20% overhead.
Final Estimate
Approximately $3 \times 10^{23}$ to $5 \times 10^{23}$ FLOPs (roughly 300 to 500 sextillion FLOPs).
For context, this is equivalent to about 3,000 to 5,000 petaflop-seconds (or PFLOP-seconds), which aligns with the estimated thousands of petaflop-days often cited for GPT-3's training run on V100 GPUs.
Vote
2 / 30
Logic Puzzlereasoning
Tests an AI's ability to solve a simple but potentially confusing logic puzzle
DeepSeek R1
No content available for this output.
Kimi K2.5
No content available for this output.
Vote
3 / 30
Stochastic Consistency Testanalysis
Tests an AI's randomness and creativity
DeepSeek R1
No content available for this output.
Kimi K2.5
Here are 5 jokes:
- Why don't skeletons fight each other? They don't have the guts.
- I would tell you a joke about time travel, but you didn't like it.
- What do you call a fake noodle? An impasta.
- Why did the bicycle fall over? It was two-tired.
- I used to play piano by ear, but now I use my hands.
5 are unique (all are distinct from one another).
Note: If "unique" means completely original and never told before, the answer is likely 0, as these follow common joke patterns. But within this set of five, none are duplicates.
Vote
4 / 30
SVG Layout Challengeimage generation
Tests an AI's ability to generate vector graphics
DeepSeek R1
No content available for this output.
Kimi K2.5
Vote
5 / 30
Xbox Controller SVG Artimage generation
Tests an AI's ability to create detailed SVG illustrations of gaming hardware
DeepSeek R1
Kimi K2.5
Vote
6 / 30
Realistic AI Interviewconversation
Tests an AI's ability to simulate personalities and predict future trends
DeepSeek R1
No content available for this output.
Kimi K2.5
No content available for this output.
Vote
7 / 30
Satirical Fake News Headlineconversation
Tests an AI's humor and understanding of current events
DeepSeek R1
No content available for this output.
Kimi K2.5
No content available for this output.
Vote
8 / 30
Character Voice Testconversation
Tests an AI's ability to write in distinct character voices
DeepSeek R1
No content available for this output.
Kimi K2.5
No content available for this output.
Vote
9 / 30
Minimalist Landing Pageweb design
Tests an AI's ability to generate a complete, working landing page
DeepSeek R1
Kimi K2.5
Vote
10 / 30
Pokémon Battle UI Recreationweb design
Recreate an interactive, nostalgic Pokémon battle UI in a single HTML file.
DeepSeek R1
Kimi K2.5
Vote
Sponsored
11 / 30
Mario Level UI Recreationweb design
Recreate an interactive, classic Mario level in a single HTML file.
DeepSeek R1
Kimi K2.5
Vote
12 / 30
Linear App Cloneweb design
Tests an AI's ability to replicate an existing UI with Tailwind CSS
DeepSeek R1
No content available for this output.
Kimi K2.5
No content available for this output.
Vote
12 of 30