[mle]/DaBench

Claude Haiku 5.5 vs Qwen3.6 35B A3B

Side A
Model page
Side B
Model page
Overall
70
vs
Overall
26

Claude Haiku 5.5 is clearly better. It wins 82 of the 87 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. Qwen3.6 35B A3B spends 3.0× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
7025.5s
3.9K, 78% thinking9.7K, 96% thinking
261m 5s
Agents
6139.5s
3.7K, 61% thinking10.5K, 91% thinking
241m 37s
Structured Extraction
9013.5s
2.6K, 83% thinking8.4K, 97% thinking
431m 3s
Coding
8630.6s
6.8K, 76% thinking12.1K, 99% thinking
31m 13s
Grounded Answers
7613.9s
2.2K, 78% thinking7.3K, 97% thinking
3349.5s
Instruction Following
9431.5s
4.3K, 82% thinking13.6K, 96% thinking
321m 25s
Business Logic
6726.0s
5.5K, 95% thinking10.1K, 99% thinking
2951.0s
Security & permissions
6723.4s
2.7K, 70% thinking6.4K, 92% thinking
4640.4s
Diagnosis & investigation
1725.5s
2.3K, 50% thinking7.5K, 89% thinking
048.4s

What you'd run

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

SpecAB
Speed
225 tok/s (better)
156 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M (better)
256K
Cost per run
$0.42 (better)
$1.47
Cost per case
$0.0028 (better)
$0.0096
Time per case
25.5s (better)
1m 5s
Quantization
—
FP8
Harness
openrouter
openrouter

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 5.5Qwen3.6 35B A3B
0255075100501001507026

Every test case

153 test cases

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

82 won by Claude Haiku 5.55 won by Qwen3.6 35B A3B66 tied