gpt-oss-120b Benchmark & Insights

Openai Cloudflare Workers AI

Updated Mar 25, 2026 All models

Sample size

40 runs

in window

Accuracy

67.5%

consensus match · 40d

Confidence

95%

over 40 runs

Window end

Mar 25, 2026

most recent run

Input price

$0.35/MTok

prompt tokens

Output price

$0.75/MTok

completion tokens

Model insights

01 The weakest full-sample performer: 10 false "unsafe" calls in just 40 days, all divergences packed into a Feb–Mar stretch where it reached for "high" as readily as "low".
02 Cheap, but a one-in-three error rate makes the savings moot.

Recent forecasts

Date

Conf.

Risk

Safe