AA-Omniscience
Artificial Analysis — AA-Omniscience accuracy (GPT-5.6 Sol max)
ev_2026_07_gpt56sol_aa_omniscience
- Tested
- 2026-07-09
- Runs
- 1
How it was measured
- Exact model
- gpt-5.6-sol
- Reasoning
- max
- Agent harness
- none
- Tool access
- AA default knowledge probe (not web-search agent)
- Prompting
- AA Omniscience protocol
PRIMARY WIS signal via AA-Omniscience. Sol max leads accuracy (~59%) but OpenAI models continue to show elevated hallucination vs Claude on this bench; Index is therefore lower than accuracy alone implies. Exact Index not scraped as a single published float on 2026-07-11 — curator maps from accuracy + known hall pattern (GPT-5.5: ~57% acc / ~86% hall) with Sol described as minor Index uplift + slightly higher hall.
Known limitations
- Exact Omniscience Index float not transcribed; normalisedScore is curator judgment.
- High accuracy with high answer rate can worsen Index vs cautious models.
- Display accuracy and hallucination narrative, not accuracy alone as "WIS".
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
AA-Omniscience
Knowledge reliability benchmark that jointly scores accuracy and hallucination / overconfidence. Rewards calibrated abstention rather than pure guess accuracy.
Benchmark source →