[mle]/DaBench

GLM 5.3 Flash vs Mellum2.1 Thinking

Side A
Model page
Side B
Model page
Overall
35
vs
Overall
9

GLM 5.3 Flash is clearly better. It wins 56 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. Mellum2.1 Thinking spends 2.6× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3555.5s
2.8K, 89% thinking10.3K, 63% thinking
91m 56s
Agents
1037.0s
607, 81% thinking10.8K, 93% thinking
92m 3s
Structured Extraction
4847.2s
2.7K, 93% thinking8.3K, 51% thinking
51m 36s
Coding
541m 52s
5.3K, 83% thinking11.3K, 38% thinking
52m 11s
Grounded Answers
3756.8s
2.9K, 93% thinking7.3K, 62% thinking
131m 24s
Instruction Following
4340.8s
2K, 77% thinking17.3K, 63% thinking
283m 10s
Business Logic
331m 18s
5.8K, 98% thinking9K, 50% thinking
01m 39s
Security & permissions
5525.2s
963, 81% thinking9.1K, 75% thinking
81m 40s
Diagnosis & investigation
027.0s
563, 63% thinking6.8K, 98% thinking
01m 15s

What you'd run

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

SpecAB
Speed
60 tok/s
90 tok/s
Machine
Hosted
Apple M5 Pro, 48 GB
Memory at 32K
—
12.9 GB
Context
1M (better)
128K
Cost per run
$0.19 (better)
$2.46 est.
Cost per case
$0.0013 (better)
$0.016 est.
Time per case
55.5s (better)
1m 56s
Quantization
FP8
Q8_0
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 FlashMellum2.1 Thinking
025507510050100150Perfect run359

Every test case

153 test cases

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

56 won by GLM 5.3 Flash5 won by Mellum2.1 Thinking92 tied