Vision / Visual Counting

Fruit Counting From Image

In the image task, the model counts bananas, apples, and oranges, even when some fruit overlap. It has to identify each fruit type and count carefully instead of giving a rough description.

Prompt

How many fruits of each type are in the image?

Bananas, apples, and oranges arranged on a wooden table.
Max tokens
100K
temperature
0
top_p
1
seed
42
presence_penalty
0
frequency_penalty
0
Reasoning effort
High
Execution
Single-shot via API

Fortytwo Prime

Fortytwo

PASS4 / 4

Based on the image, here is the count of each fruit type:
• Apples: 8
• Bananas: 7 (3 in the bunch on the left, 4 in the bunch on the right)
• Oranges: 5
Total: 20 fruits

Fortytwo identifies apples, bananas, and oranges, including the two banana bunches, and returns the exact answer-key counts.

vs
Claude Opus 4.8Anthropic
MIXED2 / 4

Here's the count of each fruit type in the image:

  • Apples: 8 (red apples in the center)
  • Oranges: 6 (scattered around the edges)
  • Bananas: 5 (2 on the left, 3 on the right)

Opus 4.8 identifies the expected fruit types and reports 8 apples, but gives 6 oranges instead of 5 and 5 bananas instead of 7.

Model verdicts

Have a complex task to evaluate?

Request a custom evaluation for your use case.

Request a demo →