[mle]/DaBench

MiniMax M3.1 Flash vs Underdog Saluki 27B

Side A
Model page
Side B
Model page
Overall
35
vs
Overall
33

Too close to call. MiniMax M3.1 Flash wins 29 of the 53 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. Underdog Saluki 27B spends 3.4× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3521.3s
2.3K, 73% thinking6.9K, 84% thinking
335m 15s
Agents
3637.0s
3.1K, 71% thinking11.8K, 97% thinking
409m 2s
Structured Extraction
487.96s
1.1K, 74% thinking4.4K, 92% thinking
433m 17s
Coding
5840.6s
4.6K, 66% thinking9.2K, 49% thinking
166m 54s
Grounded Answers
339.75s
1.2K, 71% thinking4.5K, 92% thinking
433m 22s
Instruction Following
5314.7s
1.7K, 70% thinking8K, 94% thinking
466m 5s
Business Logic
2425.9s
3.5K, 94% thinking5.2K, 81% thinking
333m 55s
Security & permissions
3012.7s
1.3K, 60% thinking4.9K, 91% thinking
433m 47s
Diagnosis & investigation
022.8s
2.1K, 51% thinking8.7K, 94% thinking
06m 39s

What you'd run

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

SpecAB
Speed
135 tok/s
23 tok/s
Machine
Hosted
Apple M5 Pro, 48 GB
Memory at 32K
—
10.2 GB
Context
1M (better)
256K
Cost per run
Not measured
Not measured
Cost per case
Not measured
Not measured
Time per case
21.3s (better)
5m 15s
Quantization
—
IQ2_XS
Harness
openrouter
llama.cpp

The race

3 lead changes

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

MiniMax M3.1 FlashUnderdog Saluki 27B
025507510050100150Perfect run3533

Every test case

153 test cases

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

29 won by MiniMax M3.1 Flash24 won by Underdog Saluki 27B100 tied