[mle]/DaBench

Beam 1.0 501B A23B vs MiniMax M3.1 Flash

Side A
Model page
Side B
Model page
Overall
57
vs
Overall
35

Beam 1.0 501B A23B is clearly better. It wins 52 of the 72 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. Beam 1.0 501B A23B spends 2.1× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
5717.2s
4K, 87% thinking2.3K, 73% thinking
3521.3s
Agents
5521.8s
4.8K, 90% thinking3.1K, 71% thinking
3637.0s
Structured Extraction
819.98s
3.4K, 91% thinking1.1K, 74% thinking
487.96s
Coding
4819.9s
4.5K, 70% thinking4.6K, 66% thinking
5840.6s
Grounded Answers
6911.6s
2.7K, 88% thinking1.2K, 71% thinking
339.75s
Instruction Following
7820.9s
5.6K, 90% thinking1.7K, 70% thinking
5314.7s
Business Logic
5714.0s
4.2K, 95% thinking3.5K, 94% thinking
2425.9s
Security & permissions
5120.6s
3.2K, 88% thinking1.3K, 60% thinking
3012.7s
Diagnosis & investigation
1724.3s
3.1K, 90% thinking2.1K, 51% thinking
022.8s

What you'd run

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

SpecAB
Speed
274 tok/s (better)
135 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M
1M
Cost per run
$0.83
Not measured
Cost per case
$0.0054
Not measured
Time per case
17.2s (better)
21.3s
Quantization
FP8
—
Harness
openrouter
openrouter

The race

5 lead changes

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

Beam 1.0 501B A23BMiniMax M3.1 Flash
025507510050100150Perfect run5735

Every test case

153 test cases

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

52 won by Beam 1.0 501B A23B20 won by MiniMax M3.1 Flash81 tied