AI Intelligence

Model performance on controlled practical tasks.

1,224 task results
3 model families
17 historical configurations

Model × reasoning effort

Recorded ability. Select a value to filter the ranking.

Recorded ability by model and reasoning effort
Effort	Luna	Sol	Terra
low	69.406	76.358	69.978
medium	70.594	77.588	74.451
high	72.665	73.219	74.602
xhigh	75.747	75.929	71.614
max	76.314	75.225	71.572
ultra	—	75.225	72.789

69.40677.588

Point estimates · — Not measured.

说明

All 17 configurations stay visible here. Color compares point estimates; overlapping intervals in the ranking do not establish a statistical winner.

Practical task results

How scoring works

Ability: calibrated task score. Strict pass: fully correct outcomes. Higher is better; lines show conditional 95% intervals.

AIQ Core 1.0.7 historical results. Rank uses calibrated ability across all 17 configurations in this dataset. Equal recorded scores share a rank.
Rank				Coverage % of tasks	Evidence	Details
10	Sol highgpt-5.6-sol	73.21959.9–83.4	25.0%18/72	100.0%72/72	Result evidence Sol · high Core 1.0.7 · Historical measurement · Rank 10 of 17 Ability 0–100 73.21959.9–83.4 · conditional 95% interval Strict pass % of scored tasks 25.0%18/72 scored tasks Gap to top score ability points -4.369 Task coverage observed / expected 72/72 Compared with Sol medium (77.588) in this dataset. Point estimates do not establish a statistical winner. Intervals & interpretation Ability interval conditional 95% 59.9–83.4 Strict pass interval 95% · % 16.4–36.1% Coverage % of tasks 100.0% The ability interval holds the item bank fixed. Historical measurements retain their original task and model versions. They do not qualify as current Core 1.1.0 results. Run identity & original JSON Model gpt-5.6-sol Recorded date 2026-08-28 Scoring version 1.0.8 Run ID `run_437b45de053bfca2ff1f69ad7dc57c006818f1ab4203786cc7db5cb6cde076ac` Open original public record ↗

Point-estimate order · intervals may overlap.

说明

Ability rank stays relative to all 17 configurations when filters change. Equal recorded ability shares a rank; point-estimate order does not establish a statistical winner.

Lines show conditional 95% intervals on the shared ability scale 54.062–88.158, including after filtering. Ability and intervals use the selected benchmark version and scoring 1.0.8. Historical measurements retain their original model and task version. Ability is specific to Core tasks and is not a human IQ norm.

Full scoring method and limitations