Prove the lift · AI model performance evaluation

llama-4-maverick vs deepseek-r1

-8-point lift for llama-4-maverick on ecommerce-catalog-norm

Test summary

This benchmark evaluates llama-4-maverick against deepseek-r1 on the ecommerce-catalog-norm v3 dataset (5 rows), scored by an LLM-as-judge (claude-haiku-4-5). llama-4-maverick reached a 26% pass rate versus 34% for deepseek-r1 — a -8-point lift.

Base model
deepseek-r1
via deepseek · 34% pass rate
Refined model
llama-4-maverick
via meta · 26% pass rate · 1.7s p95
Base pass rate
34%
Refined pass rate
26%
Refined p95
1.7s
Est. tokens saved
0pts
no lift — the base model held its ground
0%
Base model
0%
Invoked-refined
2 improved0 regressed2 unchanged

Per-row results

1Classify a listing into the right category node0%0%0
2Extract material + color from a product title0%100%+100
3Deduplicate two near-identical SKUs100%100%0
4Normalize a free-text size into a standard enum0%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 Invoked
Generated with Invoked · Browse all benchmarks