[mle]/DaBench

Qwen3.8 27B vs Claude Haiku 5.5

Side A
Model page
Side B
Model page
Overall
48
vs
Overall
70

Claude Haiku 5.5 is clearly better. It wins 64 of the 73 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. Qwen3.8 27B spends 1.8× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
4833.3s
5.9K, 91% thinking3.9K, 78% thinking
7025.5s
Agents
5736.6s
5.2K, 78% thinking3.7K, 61% thinking
6139.5s
Structured Extraction
7320.6s
4.6K, 92% thinking2.6K, 83% thinking
9013.5s
Coding
3753.0s
9.3K, 94% thinking6.8K, 76% thinking
8630.6s
Grounded Answers
6131.6s
4.2K, 91% thinking2.2K, 78% thinking
7613.9s
Instruction Following
6733.6s
6.9K, 93% thinking4.3K, 82% thinking
9431.5s
Business Logic
3632.9s
7.6K, 98% thinking5.5K, 95% thinking
6726.0s
Security & permissions
5122.6s
3.7K, 85% thinking2.7K, 70% thinking
6723.4s
Diagnosis & investigation
040.0s
5.9K, 85% thinking2.3K, 50% thinking
1725.5s

What you'd run

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

SpecAB
Speed
196 tok/s
225 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
256K
1M (better)
Cost per run
$1.28
$0.42 (better)
Cost per case
$0.0083
$0.0028 (better)
Time per case
33.3s
25.5s (better)
Quantization
FP8
—
Harness
openrouter
openrouter

The race

2 lead changes

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

Qwen3.8 27BClaude Haiku 5.5
0255075100501001504870

Every test case

153 test cases

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

9 won by Qwen3.8 27B64 won by Claude Haiku 5.580 tied