Verdict
Built from the measurements below
- Accuracy is too close to call: gpt-oss-120b scored 58.8% and gpt-oss-20b 51.2% on the same 80 questions, a gap within what chance alone could produce.
- gpt-oss-20b is 1.9× cheaper per question as billed ($0.000066 vs $0.00013), reasoning tokens included.
- gpt-oss-20b generates text 1.3× faster: 103 tokens per second against 82.
- gpt-oss-120b starts answering sooner: 0.5 s to the first token, against 0.7 s for gpt-oss-20b.
Which should you choose?
Choose gpt-oss-120b for a quicker first token and more hosts to choose from on OpenRouter (23 against 10). Choose gpt-oss-20b for the lower cost per correct answer ($0.00013 against $0.00021) and faster streaming. The widest subject gap is in science: gpt-oss-120b answered 23 of 30 correctly, gpt-oss-20b 18 of 30. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.
Accuracy gap +7.5 points (z = 0.95). We call a winner only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.