ollama pull lfm2:24b · one size measured: 24b (14 GB)
An unproven family worth trying and, so far, the least reliable model measured. It calls tools freely and finishes almost nothing.
| Size | Disk | Investigate | Operate | Analyze | Score |
|---|---|---|---|---|---|
| 24b | 14 GB | ✅ (fixed) | ❌ no-report |
❌ no-report |
1/3 |
Every failure has the same shape: tools called, readings taken, findings written as prose. On the analysis question it made eleven tool calls and composed no report.
| Workflow | Tool calls | Report |
|---|---|---|
| Investigate | 4 | ❌ |
| Operate | 3 | ❌ |
| Analyze | 11 | ❌ |
The runtime reminds such a run once, and that rescued Investigate. It did not rescue
Analyze: the reminder was delivered after nine readings and the model narrated again.
So the reminder helps models that were one call short of reporting, and lfm2 is
often further away than that.
Investigate passed in one measurement round and failed in the next with no change to
the code or the machine — the same variance seen in
mistral-small3.2, but wider. Treat any single
lfm2 result as one sample.
Not recommended for agent mode today. Kept on these pages because the failure is specific and recorded, and because a future release of the family may close it.