[mle]/DaBench

Clover 35B A3B Preview vs Qwen3.8 Flash Next

Side A
Model page
Side B
Model page
Overall
30
vs
Overall
42

Qwen3.8 Flash Next is clearly better. It wins 44 of the 62 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 2.4× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3035.1s
8.4K, 95% thinking4.1K, 80% thinking
421m 15s
Agents
2036.8s
9.6K, 91% thinking4.6K, 78% thinking
522m 4s
Structured Extraction
4327.5s
7.2K, 97% thinking3.1K, 90% thinking
5042.2s
Coding
101m 5s
12.2K, 100% thinking6.2K, 61% thinking
521m 36s
Grounded Answers
4821.0s
5.4K, 94% thinking3.1K, 88% thinking
5147.0s
Instruction Following
5332.8s
9.2K, 95% thinking5.2K, 88% thinking
621m 30s
Business Logic
1936.1s
9.8K, 99% thinking3.9K, 90% thinking
251m 0s
Security & permissions
4725.6s
5.8K, 92% thinking2.8K, 82% thinking
451m 8s
Diagnosis & investigation
037.0s
7K, 82% thinking3.4K, 69% 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 (better)
68 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
—
1M
Cost per run
Not measured
$0.46
Cost per case
Not measured
$0.003
Time per case
35.1s (better)
1m 15s
Quantization
—
—
Harness
custom
openrouter

The race

No lead changes

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

Clover 35B A3B PreviewQwen3.8 Flash Next
0255075100501001503042

Every test case

153 test cases

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

18 won by Clover 35B A3B Preview44 won by Qwen3.8 Flash Next91 tied