Updated Feb 16, 2026

Our Verdict

Qwen: Qwen3.5 397B A17BWinner

GPT-4.1Runner-up

Pick Qwen: Qwen3.5 397B A17B. In 3 blind votes, Qwen: Qwen3.5 397B A17B wins 67% of the time. That's not luck.

Clear winner

Writing DNA

Style Comparison

Similarity

98%

Qwen: Qwen3.5 397B A17B uses 2.4x more transitions

GPT-4.1

Qwen: Qwen3.5 397B A17B

58%Vocabulary54%

19wSentence Length14w

0.38Hedging0.39

9.0Bold6.9

6.1Lists4.2

0.24Emoji0.23

1.01Headings0.93

0.07Transitions0.17

Based on 25 + 23 text responses

vs

Ask them anything yourself

GPT-4.1

Qwen: Qwen3.5 397B A17B

279 AI models invented the same fake scientist.

We read every word. 250 models. 2.14 million words. This is what we found.

AI Hallucination Index 2026

Free preview13 of 58 slides

Download the free preview or get all 58 slides for $49

FAQ

Common questions

Keep going

Qwen: Qwen3.5 397B A17B vs GPT-5Cross-provider

GPT-4.1 vs Llama 4 MaverickNew provider

GPT-4.1 vs Claude Opus 4New provider

GPT-4.1vsQwen: Qwen3.5 397B A17B

38 fights queued

Why GPT-4.1?

4x more context1.0M

Why Qwen: Qwen3.5 397B A17B?

2.3x cheaper overall$0.60/M in · $3.60/M out

Leads 2 of 2 benchmarks

Stronger on SWE-bench Verified76.4% vs 54.6%

10 months newerFeb 2026

GPT-4.1Qwen: Qwen3.5 397B A17B

Input price

$2.00/M

$0.60/M

Output price

$8.00/M

$3.60/M

Context

1.0M

262K

Released

Apr 2025

Feb 2026

Benchmarks2 common

SWE-bench Verified

54.6%

76.4%+28.5%

87.4%

92.6%+5.6%

Favorites

GPT-4.1Qwen: Qwen3.5 397B A17B

Movie

Inception

Inception

Inception

Inception

Album

Abbey Road

Abbey Road

Book

Sapiens

Sapiens

Yuval Noah Harari

Frankenstein or The Modern Prometheus

Frankenstein or The Modern Prometheus

City

Kyoto

Tokyo

Game

The Legend of Zelda: Breath of the Wild

The Legend of Zelda: Breath of the Wild

Adventure, Action4.5

Tetris (1984)

Tetris (1984)

12 / 38

01

Estimate Complexityreasoning

Tests an AI's ability to make educated estimates based on technical knowledge

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

To estimate the total number of Floating Point Operations (FLOPs) required to train GPT-3, we can use the standard computational scaling laws for Transformer models. The calculation relies on three key variables: the number of parameters in the model, the number of tokens in the training dataset, and the computational cost per parameter per token.

Here is the step-by-step reasoning and calculation.

Step 1: Identify the Number of Parameters ($N$)

The defining characteristic of GPT-3 is its size. Based on the seminal paper "Language Models are Few-Shot Learners" (Brown et al., 2020), the largest version of GPT-3 has 175 billion parameters. $$N = 175 \times 10^9 = 1.75 \times 10^{11}$$

Step 2: Identify the Training Dataset Size ($D$)

The computational cost scales linearly with the amount of data the model processes. GPT-3 was trained on a diverse corpus of internet text (Common Crawl, WebText, books, Wikipedia, etc.). The authors stated the model was trained on approximately 300 billion tokens. $$D = 300 \times 10^9 = 3.0 \times 10^{11}$$

Step 3: Determine FLOPs per Parameter per Token

To calculate the total operations, we need to know how many FLOPs are performed for a single parameter when processing a single token during training. Training involves both a forward pass (to make a prediction) and a backward pass (to calculate gradients and update weights).

Forward Pass: For each parameter, the GPU performs a multiplication and an addition. This counts as 2 FLOPs per parameter.
Backward Pass: Calculating gradients is more computationally intensive. It generally requires roughly twice the operations of the forward pass to compute the gradients with respect to the activations and the weights. This counts as approximately 4 FLOPs per parameter.

Total FLOPs per parameter per token: $2 + 4 = 6$.

Note: This heuristic ($6ND$) is widely accepted in deep learning literature, specifically referenced in the "Scaling Laws for Neural Language Models" (Kaplan et al., 2020).

Step 4: Calculate Total FLOPs

Now we combine the variables into the formula: $$\text{Total FLOPs} \approx 6 \times N \times D$$

Substitute the values: $$\text{Total FLOPs} \approx 6 \times (1.75 \times 10^{11}) \times (3.0 \times 10^{11})$$

First, multiply the coefficients: $$6 \times 1.75 \times 3.0$$ $$6 \times 3.0 = 18$$ $$18 \times 1.75 = 31.5$$

Next, multiply the exponents: $$10^{11} \times 10^{11} = 10^{22}$$

Combine them: $$\text{Total FLOPs} \approx 31.5 \times 10^{22}$$

Convert to standard scientific notation: $$\text{Total FLOPs} \approx 3.15 \times 10^{23}$$

Step 5: Review and Refine

This estimate assumes a theoretically efficient implementation. In practice, there are minor overheads (embedding lookups, attention mask operations, optimizer states) that might slightly increase this number, but the $6ND$ rule captures the dominant matrix multiplications that consume the vast majority of compute.

Some analyses round the token count or parameter count slightly differently, leading to estimates ranging between $3.0 \times 10^{23}$ and $3.6 \times 10^{23}$. However, $3.15 \times 10^{23}$ is the precise result derived from the standard public specifications of the model.

Final Estimate

The total number of FLOPs required to train GPT-3 is approximately:

$$3.15 \times 10^{23} \text{ FLOPs}$$

(Often colloquially referred to as 315 zettaFLOPs).

Try this prompt

Vote

2 / 38

02

AI Board Game Logicreasoning

Tests an AI's ability to understand game rules and strategy

GPT-4.1

GPT-4.1

Openai · Apr 2025svg

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026svg

Try this prompt

Vote

3 / 38

03

Logic Puzzlereasoning

Tests an AI's ability to solve a simple but potentially confusing logic puzzle

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Sally has 1 sister.

Here is the breakdown:

Since each brother has 2 sisters, there are 2 girls in total in the family.
Sally is one of those girls.
Therefore, Sally has 1 sister (the other girl).

Try this prompt

Vote

4 / 38

04

Stochastic Consistency Testanalysis

Tests an AI's randomness and creativity

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Here are 5 jokes:

Why did the bicycle fall over? Because it was two-tired.
What do you call a fish wearing a bowtie? Sofishticated.
Why can't you give Elsa a balloon? Because she will let it go.
What do you call a lazy kangaroo? A pouch potato.
Why did the math book look sad? Because it had too many problems.

How many are unique? All 5 are unique (none of them are duplicates of each other).

Try this prompt

Vote

5 / 38

05

SVG Layout Challengeimage generation

Tests an AI's ability to generate vector graphics

GPT-4.1

GPT-4.1

Openai · Apr 2025svg

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026svg

Try this prompt

Vote

6 / 38

06

Xbox Controller SVG Artimage generation

Tests an AI's ability to create detailed SVG illustrations of gaming hardware

GPT-4.1

GPT-4.1

Openai · Apr 2025svg

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026svg

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Vote

Sponsored

7 / 38

07

Generate a Stand-Up Routineconversation

Tests an AI's humor and creative writing ability

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Vote

8 / 38

08

Realistic AI Interviewconversation

Tests an AI's ability to simulate personalities and predict future trends

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Vote

9 / 38

09

Satirical Fake News Headlineconversation

Tests an AI's humor and understanding of current events

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Vote

10 / 38

10

Character Voice Testconversation

Tests an AI's ability to write in distinct character voices

GPT-4.1

GPT-4.1

Openai · Apr 2025text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026text

Nothing here. The model returned empty. We stared at it for a while.

Try this prompt

Vote

11 / 38

11

Minimalist Landing Pageweb design

Tests an AI's ability to generate a complete, working landing page

GPT-4.1

GPT-4.1

Openai · Apr 2025website

Try this prompt

Qwen: Qwen3.5 397B A17B

Qwen: Qwen3.5 397B A17B

Qwen · Feb 2026website

Try this prompt

Vote

This matchup has more rounds

26+ more head-to-head results. Free. Not a trick.

Free account. No card required. By continuing, you agree to Rival's Terms and Privacy Policy