[mle]/DaBench

Clover 35B A3B Preview vs Spark X2.6 4B

Side A
Model page
Side B
Model page
Overall
30
vs
Overall
15

Clover 35B A3B Preview is clearly better. It wins 40 of the 51 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. The time under each score is the average per test case, on the machine it ran on. Clover 35B A3B Preview spends 1.2× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3035.1s
8.4K, 95% thinking8.2K, 84% thinking
151m 3s
Agents
2036.8s
9.6K, 91% thinking11.5K, 97% thinking
141m 32s
Structured Extraction
4327.5s
7.2K, 97% thinking5.3K, 78% thinking
1039.5s
Coding
101m 5s
12.2K, 100% thinking9.1K, 48% thinking
41m 8s
Grounded Answers
4821.0s
5.4K, 94% thinking6.6K, 86% thinking
1050.2s
Instruction Following
5332.8s
9.2K, 95% thinking11.9K, 90% thinking
371m 29s
Business Logic
1936.1s
9.8K, 99% thinking4.9K, 86% thinking
1036.3s
Security & permissions
4725.6s
5.8K, 92% thinking7K, 94% thinking
3154.3s
Diagnosis & investigation
037.0s
7K, 82% thinking11.6K, 98% thinking
01m 30s

What you'd run

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

SpecAB
Speed
252 tok/s
136 tok/s
Machine
Hosted
RTX 5090, 32 GB
Memory at 32K
—
9.4 GB
Context
—
605K
Cost per run
Not measured
$1.61 est.
Cost per case
Not measured
$0.011 est.
Time per case
35.1s (better)
1m 3s
Quantization
—
BF16
Harness
custom
llama.cpp

The race

3 lead changes

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

Clover 35B A3B PreviewSpark X2.6 4B
0255075100501001503015

Every test case

153 test cases

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

40 won by Clover 35B A3B Preview11 won by Spark X2.6 4B102 tied