Unified Memory

Benchmarks / Which Small AI Model Should You Run on a Mac? We Ran 312 Tests / Read images

vision-019

hard nutrition_labelreasoning

The prompt

1 image attached (not shown).

user

If I eat the entire container, how many grams of total sugars do I eat? Reply with only the number (no units or currency symbols).
What we expected
{
  "answer": 30,
  "accept": [],
  "type": "number"
}

What each model answered

LFM2.5-VL 3B

✗ fail · score 0 · 0.48 s

12
Scoring detail
{
  "type": "number",
  "parsed": 12
}

MiMo-V2.6 Distill 9B

✗ fail · score 0 · 1.42 s

12
Scoring detail
{
  "type": "number",
  "parsed": 12
}

MiniCPM-V 4.7 1B

✗ fail · score 0 · 0.73 s

12g
Scoring detail
{
  "type": "number",
  "parsed": 12
}

← vision-018All Read images itemsvision-020 →