Clover 35B A3B Preview vs Flint

Overall
30
vs
Overall
21

Clover 35B A3B Preview is clearly better. It wins 28 of the 41 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. Flint spends 1.2× as many thinking tokens per test case.

Thinking tokensAnswer tokens
TaskAB
Overall
3035.1s
8.4K, 95% thinking10.9K, 90% thinking
211m 17s
Agents
2036.8s
9.6K, 91% thinking4.8K, 85% thinking
1447.7s
Structured Extraction
4327.5s
7.2K, 97% thinking10.9K, 87% thinking
291m 17s
Coding
101m 5s
12.2K, 100% thinking12K, 98% thinking
141m 19s
Grounded Answers
4821.0s
5.4K, 94% thinking8.7K, 93% thinking
2957.7s
Instruction Following
5332.8s
9.2K, 95% thinking22.2K, 93% thinking
352m 28s
Business Logic
1936.1s
9.8K, 99% thinking11.7K, 78% thinking
101m 20s
Security & permissions
4725.6s
5.8K, 92% thinking7.1K, 90% thinking
3552.6s
Diagnosis & investigation
037.0s
7K, 82% thinking7.3K, 99% thinking
057.0s

What you'd run

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

SpecAB
Speed
252 tok/s (better)
149 tok/s
Machine
Hosted
Hosted
Memory at 32K
—
—
Context
—
256K
Cost per run
Not measured
Not measured
Cost per case
Not measured
Not measured
Time per case
35.1s (better)
1m 17s
Quantization
—
BF16
Harness
custom
openrouter

The race

3 lead changes

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

Clover 35B A3B PreviewFlint
0255075100501001503021

Every test case

153 test cases

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

28 won by Clover 35B A3B Preview13 won by Flint112 tied