Big Bench Extra Hard — even more challenging reasoning tasks pushing the limits of language model capabilities.
Data from LayerLens
As of November 21, 2026, the top-scoring model on BBEH is DeepSeek V4 Flash (Non-Reasoning) at 67.3%, followed by DeepSeek V4 Pro at 64.9% and GPT-5 at 64.1%. 24 models have been evaluated on this benchmark.
Last updated: November 21, 2026
Models
24
Best Score
67.3
Average
28.6
Std Dev
18.6
Provider | Model | Input $/M | Output $/M | BBEH | Actions |
|---|---|---|---|---|---|
$0.014 | $0.029 | 67.3 | |||
$0.111 | $1.322 | 64.9 | |||
$1.250 | $10.000 | 64.1 | |||
$0.125 | $1.000 | 54.9 | |||
$5.000 | $25.000 | 49.2 | |||
$5.000 | $25.000 | 49.2 | |||
$3.000 | $15.000 | 37.1 | |||
$0.025 | $0.200 | 29.3 | |||
$0.060 | $0.400 | 26.6 | |||
$0.060 | $0.400 | 26.6 | |||
$0.200 | $0.600 | 25.6 | |||
$0.100 | $0.300 | 19.5 | |||
$0.500 | $1.500 | 18.7 | |||
$2.000 | $8.000 | 17.7 | |||
$2.500 | $10.000 | 17.1 | |||
$0.800 | $3.200 | 15.9 | |||
$0.014 | $0.028 | 15.4 | |||
$0.075 | $0.200 | 14.2 | |||
$0.440 | $2.200 | 14.1 | |||
$0.340 | $0.390 | 14.0 | |||
$0.035 | $0.140 | 13.2 | |||
$1.000 | $3.000 | 11.2 | |||
$1.000 | $3.000 | 11.2 | |||
$0.070 | $0.140 | 10.1 |
Pricing from OpenRouter. Benchmarks from Artificial Analysis.
Big Bench Extra Hard — even more challenging reasoning tasks pushing the limits of language model capabilities.
This leaderboard shows all models with BBEH benchmark scores, ranked from highest to lowest. Pricing data is included to help you compare performance against cost.
Built by @aellman
Also from the founder: BlobCap, a price index for AI training data deals, showing what AI labs actually pay for data.
2026 68 Ventures, LLC. All rights reserved.