[mle]/DaBench

GPT-6 Luna vs Qwen3.8 Flash

Side A
Model page
Side B
Model page
Overall
61
vs
Overall
32

GPT-6 Luna is clearly better. It wins 70 of the 77 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. Qwen3.8 Flash spends 2.4× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
6149.1s
1.9K, 73% thinking4K, 81% thinking
3258.5s
Agents
561m 18s
1.3K, 67% thinking4.2K, 76% thinking
491m 25s
Structured Extraction
9514.5s
999, 72% thinking3K, 90% thinking
3336.9s
Coding
732m 18s
3.6K, 60% thinking5.8K, 61% thinking
471m 21s
Grounded Answers
6715.6s
1.1K, 72% thinking3.1K, 90% thinking
2641.4s
Instruction Following
8432.3s
2K, 74% thinking5K, 89% thinking
371m 4s
Business Logic
3743.1s
3.2K, 93% thinking3.9K, 91% thinking
1746.8s
Security & permissions
6224.5s
1K, 68% thinking3.2K, 83% thinking
5054.1s
Diagnosis & investigation
1737.7s
1.1K, 68% thinking3.3K, 73% thinking
058.4s

What you'd run

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

SpecAB
Speed
90 tok/s (better)
77 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1.1M (better)
1M
Cost per run
$0.10 (better)
$0.46
Cost per case
$0.00067 (better)
$0.003
Time per case
49.1s (better)
58.5s
Quantization
—
—
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.

GPT-6 LunaQwen3.8 Flash
025507510050100150Perfect run6132

Every test case

153 test cases

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

70 won by GPT-6 Luna7 won by Qwen3.8 Flash76 tied