[mle]/DaBench

Claude Haiku 5.5 vs DeepSeek V4.1 Flash

Side A
Model page
Side B
Model page
Overall
70
vs
Overall
64

Claude Haiku 5.5 is clearly better. It wins 21 of the 31 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. DeepSeek V4.1 Flash spends 1.8× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
7025.5s
3.9K, 78% thinking6K, 91% thinking
643m 35s
Agents
6139.5s
3.7K, 61% thinking6.1K, 82% thinking
464m 42s
Structured Extraction
9013.5s
2.6K, 83% thinking4.4K, 94% thinking
862m 8s
Coding
8630.6s
6.8K, 76% thinking9.7K, 95% thinking
514m 55s
Grounded Answers
7613.9s
2.2K, 78% thinking3.9K, 90% thinking
762m 17s
Instruction Following
9431.5s
4.3K, 82% thinking7.5K, 94% thinking
894m 24s
Business Logic
6726.0s
5.5K, 95% thinking7K, 97% thinking
814m 1s
Security & permissions
6723.4s
2.7K, 70% thinking4.4K, 81% thinking
693m 3s
Diagnosis & investigation
1725.5s
2.3K, 50% thinking3.2K, 67% thinking
172m 8s

What you'd run

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

SpecAB
Speed
225 tok/s (better)
30 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
1M
1M
Cost per run
$0.42
$0.39 (better)
Cost per case
$0.0028
$0.0026 (better)
Time per case
25.5s (better)
3m 35s
Quantization
—
FP8
Harness
openrouter
openrouter

The race

1 lead change

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

Claude Haiku 5.5DeepSeek V4.1 Flash
025507510050100150Perfect run7064

Every test case

153 test cases

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

21 won by Claude Haiku 5.510 won by DeepSeek V4.1 Flash122 tied