[mle]/DaBench

MiniMax M3.1 Flash vs Qwen3.8 27B

Side A
Model page
Side B
Model page
Overall
35
vs
Overall
48

Qwen3.8 27B is clearly better. It wins 48 of the 79 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 27B spends 3.2× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3521.3s
2.3K, 73% thinking5.9K, 91% thinking
4833.3s
Agents
3637.0s
3.1K, 71% thinking5.2K, 78% thinking
5736.6s
Structured Extraction
487.96s
1.1K, 74% thinking4.6K, 92% thinking
7320.6s
Coding
5840.6s
4.6K, 66% thinking9.3K, 94% thinking
3753.0s
Grounded Answers
339.75s
1.2K, 71% thinking4.2K, 91% thinking
6131.6s
Instruction Following
5314.7s
1.7K, 70% thinking6.9K, 93% thinking
6733.6s
Business Logic
2425.9s
3.5K, 94% thinking7.6K, 98% thinking
3632.9s
Security & permissions
3012.7s
1.3K, 60% thinking3.7K, 85% thinking
5122.6s
Diagnosis & investigation
022.8s
2.1K, 51% thinking5.9K, 85% thinking
040.0s

What you'd run

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

SpecAB
Speed
135 tok/s
196 tok/s (better)
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M (better)
256K
Cost per run
Not measured
$1.28
Cost per case
Not measured
$0.0083
Time per case
21.3s (better)
33.3s
Quantization
—
FP8
Harness
openrouter
openrouter

The race

1 lead change

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

MiniMax M3.1 FlashQwen3.8 27B
025507510050100150Perfect run3548

Every test case

153 test cases

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

31 won by MiniMax M3.1 Flash48 won by Qwen3.8 27B74 tied