DarkBench Leaderboard

DarkBench — benchmark testing model safety and resistance to adversarial attacks.

Data from LayerLens

As of November 21, 2026, the top-scoring model on DarkBench is Qwen3 32B at 49.2%, followed by Qwen3 32B at 49.2% and Gemini 3 Flash Preview at 49.1%. 7 models have been evaluated on this benchmark.

Last updated: November 21, 2026

8 Ways to Use Fewer Tokens

Models

7

Best Score

49.2

Average

48.0

Std Dev

2.2

Categories
Reasoning and Logic
Sponsored The Grid | Spot Priced LLM API
The Grid | Spot Priced LLM API — Why pay list price when providers will bid for your API usage? Up to 80% off.
View spot prices
Provider
Model
Input $/M
Output $/M
DarkBench
Actions
$0.064
$0.224
49.2
$0.064
$0.224
49.2
$0.500
$3.000
49.1
$0.500
$3.000
49.1
$0.550
$2.200
48.6
$0.100
$0.300
47.7
$0.800
$3.200
42.7

Pricing from OpenRouter. Benchmarks from Artificial Analysis.

About DarkBench

DarkBench — benchmark testing model safety and resistance to adversarial attacks.

This leaderboard shows all models with DarkBench benchmark scores, ranked from highest to lowest. Pricing data is included to help you compare performance against cost.

Frequently Asked Questions

DarkBench — benchmark testing model safety and resistance to adversarial attacks.
As of November 21, 2026, Qwen3 32B leads the DarkBench leaderboard with a score of 49.2. Rankings change as new models are released and evaluated.
Currently 7 models have been evaluated on DarkBench, with an average score of 48.0 and standard deviation of 2.2.
Benchmark scores are updated when new evaluations are published by our data sources (Artificial Analysis and LayerLens). Pricing data is refreshed daily from OpenRouter.