[mle]/DaBench

Claude Haiku 5.5 vs Clover 35B A3B Preview

Side A
Model page
Side B
Model page
Overall
70
vs
Overall
30

Claude Haiku 5.5 is clearly better. It wins 77 of the 82 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 2.6× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
7025.5s
3.9K, 78% thinking8.4K, 95% thinking
3035.1s
Agents
6139.5s
3.7K, 61% thinking9.6K, 91% thinking
2036.8s
Structured Extraction
9013.5s
2.6K, 83% thinking7.2K, 97% thinking
4327.5s
Coding
8630.6s
6.8K, 76% thinking12.2K, 100% thinking
101m 5s
Grounded Answers
7613.9s
2.2K, 78% thinking5.4K, 94% thinking
4821.0s
Instruction Following
9431.5s
4.3K, 82% thinking9.2K, 95% thinking
5332.8s
Business Logic
6726.0s
5.5K, 95% thinking9.8K, 99% thinking
1936.1s
Security & permissions
6723.4s
2.7K, 70% thinking5.8K, 92% thinking
4725.6s
Diagnosis & investigation
1725.5s
2.3K, 50% thinking7K, 82% thinking
037.0s

What you'd run

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

SpecAB
Speed
225 tok/s
252 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M
—
Cost per run
$0.42
Not measured
Cost per case
$0.0028
Not measured
Time per case
25.5s (better)
35.1s
Quantization
—
—
Harness
openrouter
custom

The race

No lead changes

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

Claude Haiku 5.5Clover 35B A3B Preview
0255075100501001507030

Every test case

153 test cases

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

77 won by Claude Haiku 5.55 won by Clover 35B A3B Preview71 tied