[mle]/DaBench

GPT-6 Luna vs Mistral Large 4

Side A
Model page
Side B
Model page
Overall
61
vs
Overall
28

GPT-6 Luna is clearly better. It wins 64 of the 69 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. Mistral Large 4 spends 3.2× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
6149.1s
1.9K, 73% thinking6.2K, 72% thinking
281m 56s
Agents
561m 18s
1.3K, 67% thinking795, no thinking
1126.1s
Structured Extraction
9514.5s
999, 72% thinking288, no thinking
153.78s
Coding
732m 18s
3.6K, 60% thinking9.3K, 37% thinking
332m 48s
Grounded Answers
6715.6s
1.1K, 72% thinking8.1K, 89% thinking
382m 45s
Instruction Following
8432.3s
2K, 74% thinking12.9K, 84% thinking
693m 47s
Business Logic
3743.1s
3.2K, 93% thinking8.8K, 77% thinking
242m 40s
Security & permissions
6224.5s
1K, 68% thinking3.7K, 83% thinking
331m 9s
Diagnosis & investigation
1737.7s
1.1K, 68% thinking4.4K, 85% thinking
01m 19s

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)
56 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1.1M (better)
512K
Cost per run
$0.10 (better)
$2.67
Cost per case
$0.00067 (better)
$0.017
Time per case
49.1s (better)
1m 56s
Quantization
—
—
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.

GPT-6 LunaMistral Large 4
025507510050100150Perfect run6128

Every test case

153 test cases

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

64 won by GPT-6 Luna5 won by Mistral Large 484 tied