Model comparisons
Same data, same protocol, verified pricing on both sides. Where the evidence does not decide, the page says so.
Head-to-head runs
Banking77 · n=208 · pre-registeredJev vs GPT-5.6 TerraTerra leads 87.5% to 83.2% on accuracy, but the intervals overlap — the run's own verdict is AMBIGUOUS. Terra answers in 1510ms against Jev's 440ms and lists input at 48x Jev's price.READ →Banking77 · n=208 · pre-registeredJev vs GPT-5.4 NanoJev leads 83.2% to 79.3% on accuracy with overlapping intervals — no proven winner. Jev answers in 440ms against 850ms; reaching a 90.5% accuracy target needs fallback 22% of the time for Jev, 48.5% for Nano.READ →2,000-email phishing benchmarkJev vs Claude Haiku 4.5Jev 1.13 against Claude Haiku 4.5 on a reproducible phishing-email set: accuracy, recall, calibration, latency and cost.READ →
A comparison is published only when both models ran head-to-head on the same data under the same protocol and both list prices are verified. Pages are not generated per model name.