The publisher’s hallucination-rate measure for AA-Omniscience v1.1; lower values indicate fewer unsupported answers on this scale.
Score version
28 model results
One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
AA-Omniscience v1.1 · hallucination rate, lower is better · 2026-10-28
Publisher-selected recent-model cohort from the public evaluation page. Model labels preserve effort and fallback configurations. Values are a dated snapshot, not a live feed. Consult the publisher’s methodology for the harness, task split and judging; do not merge this cohort with vendor-run results.
AA-Omniscience · hallucination rate, lower is better · 2026-10-26
Publisher-selected recent-model cohort from the public evaluation page. Model labels preserve effort and fallback configurations. Values are a dated snapshot, not a live feed. Consult the publisher’s methodology for the harness, task split and judging; do not merge this cohort with vendor-run results. Archived RLTasks snapshot from 2026-10-26. The publisher URL now displays a newer version; these original observations retain the earlier evaluation contract.
TRACE THE SCORE · Per-model notes use published breakdowns where available.
Reading the result
Keep the publisher’s denominator and partial-answer policy. Do not substitute one minus accuracy for hallucination rate.
Version boundary
The refreshed question set is not directly comparable with v1. Version buttons keep older observations available without averaging across datasets.
Where this fits in an evaluation suite
Read accuracy together with error and abstention rates. Domain averages can hide language gaps, specialist failures, and poorly calibrated confidence.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
Artificial Analysis v1.1: 3,000 questions, one repeat, no tools. Retain v1’s 6,000-question results as a distinct version.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.