[mle]/DaBench

DeepSeek V4.1 Flash 1017 vs Qwen3.8 27B

Side A
Model page
Side B
Model page
Overall
56
vs
Overall
48

DeepSeek V4.1 Flash 1017 is clearly better. It wins 47 of the 64 test cases where they differ.

Learn to build evals like this in the AI Engineering Academy

Task by task

A score turns orange only when that model wins clearly more of the task's test cases. Under each score, the average time per test case on the machine it ran on. Both models spend about the same on thinking per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
5626.0s
5.3K, 91% thinking5.9K, 91% thinking
4833.3s
Agents
5826.7s
3.3K, 71% thinking5.2K, 78% thinking
5736.6s
Structured Extraction
8117.2s
4.2K, 93% thinking4.6K, 92% thinking
7320.6s
Coding
4840.8s
9.4K, 95% thinking9.3K, 94% thinking
3753.0s
Grounded Answers
7119.8s
4.4K, 91% thinking4.2K, 91% thinking
6131.6s
Instruction Following
7625.6s
5.5K, 91% thinking6.9K, 93% thinking
6733.6s
Business Logic
6726.6s
6.6K, 97% thinking7.6K, 98% thinking
3632.9s
Security & permissions
4826.6s
4.7K, 87% thinking3.7K, 85% thinking
5122.6s
Diagnosis & investigation
020.2s
3.2K, 74% thinking5.9K, 85% thinking
040.0s

What you'd run

The build behind each score. Orange marks the faster, smaller, cheaper or longer-context side.

SpecAB
Speed
243 tok/s (better)
196 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M (better)
256K
Cost per run
$0.75 (better)
$1.28
Cost per case
$0.0049 (better)
$0.0083
Time per case
26.0s (better)
33.3s
Quantization
—
FP8
Harness
openrouter
openrouter

The race

1 lead change

Each model's running score as the test cases go by. The dashed line is a perfect run.

DeepSeek V4.1 Flash 1017Qwen3.8 27B
025507510050100150Perfect run5648

Every test case

153 test cases

Each square is one test case, colored by the model that scored higher.

47 won by DeepSeek V4.1 Flash 101717 won by Qwen3.8 27B89 tied