Prove the lift · AI model performance evaluation
nemotron-ultra-253b vs deepseek-v3
+19-point lift for nemotron-ultra-253b on contract-clause-qa
Test summary
This benchmark evaluates nemotron-ultra-253b against deepseek-v3 on the contract-clause-qa v2 dataset (5 rows), scored by an LLM-as-judge (claude-haiku-4-5, run locally). nemotron-ultra-253b reached a 44% pass rate versus 25% for deepseek-v3 — a +19-point lift.
- Base model
- deepseek-v3
- via deepseek · 25% pass rate
- Refined model
- nemotron-ultra-253b
- via nvidia · 44% pass rate · 2.8s p95
Base pass rate
25%
Refined pass rate
44%
Refined p95
2.8s
Est. tokens saved
6.5K
+0pts
refined beats base on 5 rows
0%
Base model0%
Invoked-refined▲1 improved▼0 regressed●4 unchanged
Per-row results
1Find the liability cap and its exceptions100% → 100%0
2Identify the governing-law clause and jurisdiction0% → 0%0
3Extract the auto-renewal notice window in days0% → 0%0
4Locate the termination-for-convenience terms100% → 100%0
5Determine indemnification scope and carve-outs0% → 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