TokenJam proof — judged

tokenjam 0.5.2   n=5 tasks · k=1 sample(s) · openai:gpt-4o → openai:gpt-4o-mini

Not enough evidence yet

We ran the same 5 judged task(s) through both models and graded every answer with the same automated judge. The cheaper candidate (gpt-4o-mini) cost 93% less than gpt-4o, and scored 20 points higher on this suite — 40% of tasks passed before, 60% after.

Too few tasks were run to be statistically sure either way. Run more before deciding.

-93%
cheaper to run (measured API $)
+20 pts
accuracy vs the original model
0 worse · 1 better
tasks changed by the swap (of 5)

How each task did

TaskOriginalCheaper modelWhy — the judge’s reason
Refund policyfailfailThe actual output provides a general explanation of what a refund window is and graded 0.30 (needs 0.50)
CapitalpasspassThe actual output matches the expected output exactly in terms of factual inform graded 1.00 (needs 0.50)
Retry summarypasspassThe actual output accurately reflects the factual information of the expected ou graded 0.90 (needs 0.50)
ShippingfailfailThe actual output provides a detailed breakdown of standard shipping times based graded 0.30 (needs 0.50)
Define llmfailpassThe actual output provides a broader description of a large language model, focu graded 0.50 (needs 0.50)

The statistics behind it

Verdict: Insufficient evidence  ·  McNemar p=1.000 (α=0.05)  ·  candidate chosen by explicit --candidate override

McNemar’s test asks whether the difference between the two models is bigger than chance: a p-value above 0.05 means the change is not statistically significant. The 95% CI on the pass-rate delta is the range the true difference is likely in — if it crosses zero, the direction isn’t certain.

+20.0pp
pass-rate delta [95% CI -15.1, +55.1]

Pass rate (95% CI whiskers)

Original2/5 (40%)
Candidate3/5 (60%)

Cost (measured)

Original$0.002610
Candidate$0.000182

How to read this

Generated 2026-06-26 13:42 · tokenjam-bench · proof report