Benchmarks / Which Small AI Model Should You Run on a Mac? We Ran 312 Tests / Jobs
vision · 40 items
Read images
Read receipts, screenshots, signs, labels, charts, calendars and handwriting from a photo or screenshot, and answer with the exact value.
Results
| Model | Accuracy | Pass | Easy | Medium | Hard | Median latency | Tok/s | Job time |
|---|---|---|---|---|---|---|---|---|
| LFM2.5-VL 3B | 85.0 | 85.0% | 100.0% | 80.0% | 76.9% | 2.45 s | 114.3 | 1m 22s |
| MiMo-V2.6 Distill 9B | 77.5 | 77.5% | 91.7% | 73.3% | 69.2% | 1.99 s | 43.1 | 1m 26s |
| MiniCPM-V 4.7 1B | 72.5 | 72.5% | 100.0% | 66.7% | 53.8% | 1.19 s | 199.5 | 45.0 s |
How it is scored
Exact match after normalization, by answer type: numbers (only one distinct number in the reply, tolerance 0.005), times (one time, compared in 24h), yes/no (first word), text (casefold/punctuation-insensitive; codes and phone numbers compared as letters+digits or digits only), choices (exact label), JSON (fraction of fields correct). Pass = score 1.0.
Temperature 0 · max 1024 output tokens.
System prompt every model got
You read images for the user. Look carefully at the attached image and answer the question about it. Give exactly the answer format the user asks for, with no explanation.
Every item
Final score per item (0-100). Click any row to see the prompt, the expected answer and every model's reply. Columns sort.
✗ marks a fail (below the job's pass bar). A dash means no valid result.