[mle]/DaBench

GPT-6 Luna vs Claude Haiku 5.5

Side A
Model page
Side B
Model page
Overall
61
vs
Overall
70

Claude Haiku 5.5 is clearly better. It wins 28 of the 37 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. Under each score, the average time per test case on the machine it ran on. Claude Haiku 5.5 spends 2.2× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
6149.1s
1.9K, 73% thinking3.9K, 78% thinking
7025.5s
Agents
561m 18s
1.3K, 67% thinking3.7K, 61% thinking
6139.5s
Structured Extraction
9514.5s
999, 72% thinking2.6K, 83% thinking
9013.5s
Coding
732m 18s
3.6K, 60% thinking6.8K, 76% thinking
8630.6s
Grounded Answers
6715.6s
1.1K, 72% thinking2.2K, 78% thinking
7613.9s
Instruction Following
8432.3s
2K, 74% thinking4.3K, 82% thinking
9431.5s
Business Logic
3743.1s
3.2K, 93% thinking5.5K, 95% thinking
6726.0s
Security & permissions
6224.5s
1K, 68% thinking2.7K, 70% thinking
6723.4s
Diagnosis & investigation
1737.7s
1.1K, 68% 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
90 tok/s
225 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1.1M (better)
1M
Cost per run
$0.10 (better)
$0.42
Cost per case
$0.00067 (better)
$0.0028
Time per case
49.1s
25.5s (better)
Quantization
—
—
Harness
openrouter
openrouter

The race

5 lead changes

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

GPT-6 LunaClaude Haiku 5.5
025507510050100150Perfect run6170

Every test case

153 test cases

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

9 won by GPT-6 Luna28 won by Claude Haiku 5.5116 tied