An independent test of 9,000 calls: slower, 2× the price, no more accurate — but it sees images

Independent researchers at Nexus Agent ran more than 9,000 timed calls against both APIs during the Decisions beta: the medians tied (~268 ms vs 269 ms), but OpenAI's tail was 2.8× slower at the 99th percentile, with 95 calls over a second versus Jev's 3. On price, OpenAI ran about 2.1× more per call ($0.10/M input vs $0.042), and accuracy was a statistical tie — 233 errors against 221 across 2,155 labeled cases.

Where OpenAI won: images. Jev reads text only, while the Decisions API reads screenshots, scans, and uploads — a real edge for browser agents and image moderation. Nexus also flagged its own conflict of interest: it uses Jev in its own assistant, and asks readers to read the numbers accordingly.