[mle]/DaBench

GLM 5.3 Flash vs Spark X2.6 4B

Side A
Model page
Side B
Model page
Overall
35
vs
Overall
15

GLM 5.3 Flash is clearly better. It wins 52 of the 61 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. Spark X2.6 4B spends 2.8× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3555.5s
2.8K, 89% thinking8.2K, 84% thinking
151m 3s
Agents
1037.0s
607, 81% thinking11.5K, 97% thinking
141m 32s
Structured Extraction
4847.2s
2.7K, 93% thinking5.3K, 78% thinking
1039.5s
Coding
541m 52s
5.3K, 83% thinking9.1K, 48% thinking
41m 8s
Grounded Answers
3756.8s
2.9K, 93% thinking6.6K, 86% thinking
1050.2s
Instruction Following
4340.8s
2K, 77% thinking11.9K, 90% thinking
371m 29s
Business Logic
331m 18s
5.8K, 98% thinking4.9K, 86% thinking
1036.3s
Security & permissions
5525.2s
963, 81% thinking7K, 94% thinking
3154.3s
Diagnosis & investigation
027.0s
563, 63% thinking11.6K, 98% thinking
01m 30s

What you'd run

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

SpecAB
Speed
60 tok/s
136 tok/s
Machine
Hosted
RTX 5090, 32 GB
Memory at 32K
—
9.4 GB
Context
1M (better)
605K
Cost per run
$0.19 (better)
$1.61 est.
Cost per case
$0.0013 (better)
$0.011 est.
Time per case
55.5s (better)
1m 3s
Quantization
FP8
BF16
Harness
openrouter
llama.cpp

The race

2 lead changes

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

GLM 5.3 FlashSpark X2.6 4B
0255075100501001503515

Every test case

153 test cases

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

52 won by GLM 5.3 Flash9 won by Spark X2.6 4B92 tied