Workflows / call-summary-qa · Compare · compared 20 Oct, 3:30 pm by ops.member

Compare models

No candidate passesNo model passedIndicative only
Request a fine-tune

Gold v1Rubric v1quality bar ≥ 85%p99 ≤ 15,000 ms 180 gold rows replayed.

Results. One row per model, scored on your gold set.
ModelQualityp99 latencyCost per requestvs todayResult
Gemini 4 Argon Current modelopenrouter/google/gemini-4-argon81.1%
replayed on 180 gold rows
24,557 ms
benchmark replies on the same messages
₹0.52
benchmark replies on the same messages
—Today
openai_compatible/dvar:gemma-4-31b-it34.4%
replayed on 180 gold rows
31,207 ms
replayed on 180 gold rows
₹0.0412
replayed on 180 gold rows
−92%Fails 4 checks
openai_compatible/dvar:gemma-4-31b-it_8-examples71.7%
replayed on 180 gold rows
21,383 ms
replayed on 180 gold rows
₹0.0696
replayed on 180 gold rows
−87%Fails 3 checks
No candidate here is fine-tuned. “+ 8 examples” is the open model with 8 reviewed summaries of other calls in its prompt. A model tuned on reviewed, masked calls is the next step, through this same gate: Fine-tune.

No candidate passes every check. Open a model's row for its reasons.

Prices are per request, at each provider's list price.

Judge: 0 of 8 checks trusted.

Score by topic. Score for each topic, today against each candidate.Scale 0–1minimum per topicbelow minimum
TopicRowsToday · Gemini 4 ArgonCandidate · dvar:gemma-4-31b-itCandidate · dvar:gemma-4-31b-it_8-examples
Term life sales30
0.33
0.00
−0.33
0.67
+0.34
Health sales and porting30
1.00
0.40
−0.60
0.67
−0.33
Savings and ULIP sales30
0.97
0.40
−0.57
0.87
−0.10
Motor renewals30
1.00
0.47
−0.53
1.00
0.00
Claims support30
0.63
0.33
−0.30
0.33
−0.30
Policy service30
0.93
0.47
−0.46
0.77
−0.16
All topics, weighted by traffic180
0.81
0.34
−0.47
0.72
−0.09
——