[mle]/DaBench

Mistral Large 4 vs Qwen3.8 Flash

Side A
Model page
Side B
Model page
Overall
28
vs
Overall
32

Too close to call. Qwen3.8 Flash wins 33 of the 60 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 1.4× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
281m 56s
6.2K, 72% thinking4K, 81% thinking
3258.5s
Agents
1126.1s
795, no thinking4.2K, 76% thinking
491m 25s
Structured Extraction
153.78s
288, no thinking3K, 90% thinking
3336.9s
Coding
332m 48s
9.3K, 37% thinking5.8K, 61% thinking
471m 21s
Grounded Answers
382m 45s
8.1K, 89% thinking3.1K, 90% thinking
2641.4s
Instruction Following
693m 47s
12.9K, 84% thinking5K, 89% thinking
371m 4s
Business Logic
242m 40s
8.8K, 77% thinking3.9K, 91% thinking
1746.8s
Security & permissions
331m 9s
3.7K, 83% thinking3.2K, 83% thinking
5054.1s
Diagnosis & investigation
01m 19s
4.4K, 85% 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
56 tok/s
77 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
512K
1M (better)
Cost per run
$2.67
$0.46 (better)
Cost per case
$0.017
$0.003 (better)
Time per case
1m 56s
58.5s (better)
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.

Mistral Large 4Qwen3.8 Flash
025507510050100150Perfect run2832

Every test case

153 test cases

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

27 won by Mistral Large 433 won by Qwen3.8 Flash93 tied