Benchmarks / Which Small AI Model Should You Run on a Mac? We Ran 312 Tests / Jobs
chat · 44 items
Summarize + extract
Pull exact JSON out of messy receipts, invoices, invites, shipping emails and signatures, and summarize emails, transcripts and articles.
Results
| Model | Accuracy | Pass | Easy | Medium | Hard | Median latency | Tok/s | Job time |
|---|---|---|---|---|---|---|---|---|
| MiMo-V2.6 Distill 9B | 96.1 | 75.0% | 80.0% | 76.9% | 68.8% | 3.62 s | 37.6 | 2m 55s |
| Mellum2.1 12B-A2.5B | 92.2 | 56.8% | 73.3% | 46.2% | 50.0% | 1.62 s | 101.2 | 1m 18s |
| Ornith-1.5 9B | 89.9 | 59.1% | 66.7% | 76.9% | 37.5% | 3.27 s | 46.1 | 2m 36s |
| Granite 4.2 8B | 89.5 | 45.5% | 60.0% | 38.5% | 37.5% | 3.70 s | 43.8 | 3m 24s |
| MiniCPM5 2B | 88.8 | 47.7% | 60.0% | 23.1% | 56.2% | 1.32 s | 98.3 | 58.5 s |
| Spark-X2.5 4B | 88.0 | 54.5% | 66.7% | 53.8% | 43.8% | 2.35 s | 65.1 | 1m 51s |
| Ling-3.0-tiny (7.9B-A1.3B) | 87.1 | 47.7% | 53.3% | 46.2% | 43.8% | 1.07 s | 116.3 | 48.4 s |
| K2-Horizon 0.9B | 74.9 | 20.5% | 26.7% | 23.1% | 12.5% | 0.77 s | 210.2 | 46.5 s |
How it is scored
Extraction: per-field exact match after normalization (numbers +-0.005, phones digits-only, emails case-insensitive, null must be null, lists order-insensitive); score = fraction of fields correct, pass = all correct, invalid JSON = 0. Summaries: deterministic partial score = key-fact coverage x (0.7 + 0.15 word-limit + 0.15 no-invented-numbers), plus a judge (rubric.md) for fluency and faithfulness.
Temperature 0 · max 1024 output tokens.
System prompt every model got
You are a careful assistant that extracts information from documents and summarizes text. Follow the requested output format exactly. Use only information stated in the text.
Every item
Final score per item (0-100), which for this job blends the scripted checks with the judge panel's mark, as in the published results. 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.