Prove the lift · AI model performance evaluation
claude-sonnet-4-5 vs gpt-oss-20b
+24-point lift for claude-sonnet-4-5 on support-ticket-triage
Test summary
This benchmark evaluates claude-sonnet-4-5 against gpt-oss-20b on the support-ticket-triage v2 dataset (6 rows), scored by an LLM-as-judge (gpt-4o-mini). claude-sonnet-4-5 reached a 57% pass rate versus 33% for gpt-oss-20b — a +24-point lift.
- Base model
- gpt-oss-20b
- via openai · 33% pass rate
- Refined model
- claude-sonnet-4-5
- via anthropic · 57% pass rate · 2.6s p95
Base pass rate
33%
Refined pass rate
57%
Refined p95
2.6s
Est. tokens saved
6.8K
+0pts
refined beats base on 6 rows
0%
Base model0%
Invoked-refined▲3 improved▼0 regressed●3 unchanged
Per-row results
1Classify: billing dispute vs refund request vs cancellation100% → 100%0
2Route a multi-part ticket to the correct queue0% → 0%0
3Summarize a 12-message thread into a one-line disposition0% → 100%+100
4Detect churn-risk sentiment in a renewal thread0% → 100%+100
5Extract the affected order id from a free-text complaint0% → 0%0
6Priority-tag an outage report vs a feature question0% → 100%+100
Prove the lift on your own data.
Run any two models over your tasks, judged automatically, and see the improvement — in seconds.
Try InvokedGenerated with Invoked · Browse all benchmarks