[mle]/DaBench

DeepSeek V4.1 Flash vs Muse Spark 1.3 Contributor

Side A
Model page
Side B
Model page
Overall
64
vs
Overall
66

Too close to call. Muse Spark 1.3 Contributor wins 18 of the 30 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. Muse Spark 1.3 Contributor spends 1.5× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
643m 35s
6K, 91% thinking8.7K, 91% thinking
661m 42s
Agents
464m 42s
6.1K, 82% thinking14.7K, 88% thinking
573m 16s
Structured Extraction
862m 8s
4.4K, 94% thinking4.2K, 91% thinking
9037.9s
Coding
514m 55s
9.7K, 95% thinking10.8K, 95% thinking
482m 1s
Grounded Answers
762m 17s
3.9K, 90% thinking6.1K, 91% thinking
811m 1s
Instruction Following
894m 24s
7.5K, 94% thinking8.7K, 91% thinking
841m 18s
Business Logic
814m 1s
7K, 97% thinking6K, 95% thinking
761m 7s
Security & permissions
693m 3s
4.4K, 81% thinking8.2K, 87% thinking
761m 59s
Diagnosis & investigation
172m 8s
3.2K, 67% thinking17.2K, 90% thinking
173m 34s

What you'd run

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

SpecAB
Speed
30 tok/s
107 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M
1M
Cost per run
$0.39 (better)
$0.41
Cost per case
$0.0026 (better)
$0.0027
Time per case
3m 35s
1m 42s (better)
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 FlashMuse Spark 1.3 Contributor
025507510050100150Perfect run6466

Every test case

153 test cases

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

12 won by DeepSeek V4.1 Flash18 won by Muse Spark 1.3 Contributor123 tied