[mle]/DaBench

Claude Haiku 5.5 vs Spark X2.6 1.7B

Side A
Model page
Side B
Model page
Overall
70
vs
Overall
4

Claude Haiku 5.5 is clearly better. It wins 113 of the 114 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. The time under each score is the average per test case, on the machine it ran on. Spark X2.6 1.7B spends 2.4× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
7025.5s
3.9K, 78% thinking10.8K, 68% thinking
41m 43s
Agents
6139.5s
3.7K, 61% thinking13.6K, 95% thinking
02m 10s
Structured Extraction
9013.5s
2.6K, 83% thinking7.3K, 57% thinking
01m 19s
Coding
8630.6s
6.8K, 76% thinking11.6K, 37% thinking
01m 51s
Grounded Answers
7613.9s
2.2K, 78% thinking6.1K, 66% thinking
056.4s
Instruction Following
9431.5s
4.3K, 82% thinking18.7K, 72% thinking
242m 53s
Business Logic
6726.0s
5.5K, 95% thinking8.9K, 49% thinking
01m 22s
Security & permissions
6723.4s
2.7K, 70% thinking6.5K, 70% thinking
1059.4s
Diagnosis & investigation
1725.5s
2.3K, 50% thinking21.5K, 96% thinking
03m 16s

What you'd run

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

SpecAB
Speed
225 tok/s
106 tok/s
Machine
Hosted
Apple M5 Pro, 48 GB
Memory at 32K
—
2.9 GB
Context
1M
1M
Cost per run
$0.42
Not measured
Cost per case
$0.0028
Not measured
Time per case
25.5s (better)
1m 43s
Quantization
—
Q8_0
Harness
openrouter
llama.cpp

The race

No lead changes

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

Claude Haiku 5.5Spark X2.6 1.7B
025507510050100150704

Every test case

153 test cases

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

113 won by Claude Haiku 5.51 won by Spark X2.6 1.7B39 tied