[mle]/DaBench

Clover 35B A3B Preview vs Mellum2.1 Thinking

Side A
Model page
Side B
Model page
Overall
30
vs
Overall
9

Clover 35B A3B Preview is clearly better. It wins 45 of the 52 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% thinking10.3K, 63% thinking
91m 56s
Agents
2036.8s
9.6K, 91% thinking10.8K, 93% thinking
92m 3s
Structured Extraction
4327.5s
7.2K, 97% thinking8.3K, 51% thinking
51m 36s
Coding
101m 5s
12.2K, 100% thinking11.3K, 38% thinking
52m 11s
Grounded Answers
4821.0s
5.4K, 94% thinking7.3K, 62% thinking
131m 24s
Instruction Following
5332.8s
9.2K, 95% thinking17.3K, 63% thinking
283m 10s
Business Logic
1936.1s
9.8K, 99% thinking9K, 50% thinking
01m 39s
Security & permissions
4725.6s
5.8K, 92% thinking9.1K, 75% thinking
81m 40s
Diagnosis & investigation
037.0s
7K, 82% thinking6.8K, 98% thinking
01m 15s

What you'd run

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

SpecAB
Speed
252 tok/s
90 tok/s
Machine
Hosted
Apple M5 Pro, 48 GB
Memory at 32K
—
12.9 GB
Context
—
128K
Cost per run
Not measured
$2.46 est.
Cost per case
Not measured
$0.016 est.
Time per case
35.1s (better)
1m 56s
Quantization
—
Q8_0
Harness
custom
llama.cpp

The race

2 lead changes

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

Clover 35B A3B PreviewMellum2.1 Thinking
025507510050100150309

Every test case

153 test cases

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

45 won by Clover 35B A3B Preview7 won by Mellum2.1 Thinking101 tied