Prove the lift · AI model performance evaluation
gpt-4.1 vs claude-haiku-4-5
+17-point lift for gpt-4.1 on support-ticket-triage
Test summary
This benchmark evaluates gpt-4.1 against claude-haiku-4-5 on the support-ticket-triage v3 dataset (6 rows), scored by an LLM-as-judge (claude-haiku-4-5). gpt-4.1 reached a 73% pass rate versus 56% for claude-haiku-4-5 — a +17-point lift.
- Base model
- claude-haiku-4-5
- via anthropic · 56% pass rate
- Refined model
- gpt-4.1
- via openai · 73% pass rate · 1.7s p95
Base pass rate
56%
Refined pass rate
73%
Refined p95
1.7s
Est. tokens saved
8.0K
+0pts
refined beats base on 6 rows
0%
Base model0%
Invoked-refined▲3 improved▼0 regressed●3 unchanged
Per-row results
1Summarize a 12-message thread into a one-line disposition100% → 100%0
2Extract the affected order id from a free-text complaint100% → 100%0
3Route a multi-part ticket to the correct queue0% → 100%+100
4Detect churn-risk sentiment in a renewal thread0% → 100%+100
5Classify: billing dispute vs refund request vs cancellation100% → 100%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