Prove the lift · AI model performance evaluation
deepseek-v3 vs gpt-4.1
+8-point lift for deepseek-v3 on medical-coding-extract
Test summary
This benchmark evaluates deepseek-v3 against gpt-4.1 on the medical-coding-extract v3 dataset (5 rows), scored by an LLM-as-judge (llama-3.1-8b). deepseek-v3 reached a 58% pass rate versus 50% for gpt-4.1 — a +8-point lift.
- Base model
- gpt-4.1
- via openai · 50% pass rate
- Refined model
- deepseek-v3
- via deepseek · 58% pass rate · 3.2s p95
Base pass rate
50%
Refined pass rate
58%
Refined p95
3.2s
Est. tokens saved
4.9K
+0pts
refined beats base on 5 rows
0%
Base model0%
Invoked-refined▲2 improved▼0 regressed●3 unchanged
Per-row results
1Map a discharge summary to the primary ICD-10 code0% → 100%+100
2Identify a contraindication from a problem list100% → 100%0
3Extract medication + dosage from a clinical note0% → 100%+100
4Determine the correct E/M level from a visit note100% → 100%0
5Pull the procedure code from an operative report100% → 100%0
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