[mle]/DaBench

Flint vs GLM 5.3 Flash

Side A
Model page
Side B
Model page
Overall
21
vs
Overall
35

GLM 5.3 Flash is clearly better. It wins 38 of the 54 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. Flint spends 4.0× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
211m 17s
10.9K, 90% thinking2.8K, 89% thinking
3555.5s
Agents
1447.7s
4.8K, 85% thinking607, 81% thinking
1037.0s
Structured Extraction
291m 17s
10.9K, 87% thinking2.7K, 93% thinking
4847.2s
Coding
141m 19s
12K, 98% thinking5.3K, 83% thinking
541m 52s
Grounded Answers
2957.7s
8.7K, 93% thinking2.9K, 93% thinking
3756.8s
Instruction Following
352m 28s
22.2K, 93% thinking2K, 77% thinking
4340.8s
Business Logic
101m 20s
11.7K, 78% thinking5.8K, 98% thinking
331m 18s
Security & permissions
3552.6s
7.1K, 90% thinking963, 81% thinking
5525.2s
Diagnosis & investigation
057.0s
7.3K, 99% thinking563, 63% thinking
027.0s

What you'd run

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

SpecAB
Speed
149 tok/s (better)
60 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
256K
1M (better)
Cost per run
Not measured
$0.19
Cost per case
Not measured
$0.0013
Time per case
1m 17s
55.5s (better)
Quantization
BF16
FP8
Harness
openrouter
openrouter

The race

2 lead changes

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

FlintGLM 5.3 Flash
0255075100501001502135

Every test case

153 test cases

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

16 won by Flint38 won by GLM 5.3 Flash99 tied