GDPval-AA v2
Long-horizon professional knowledge work with tools, files, and deliverables.
GDPval-AA v2; human expert baseline 1000
ACTIVE MEASURE
GDPval-AA v2
Long-horizon professional knowledge work with tools, files, and deliverables. Methodology ↗
RANKMODELACCESSSCORERELATIVE
01Claude Opus 5AnthropicProprietary1861Elo↗02Claude Fable 5AnthropicProprietary1760Elo↗03Kimi K3Moonshot AIOpen weights1668Elo↗04Muse Spark 1.2MetaProprietary1631Elo↗05Grok 4.5SpaceXAIProprietary1543Elo↗06Kimi K2.6Moonshot AIOpen weights1520Elo↗07GPT-5.5OpenAIProprietary1494Elo↗08North Mini CodeCohereOpen weights21.7Elo↗Ranks apply only to the selected evaluation and cited snapshot. Missing scores are not estimated. Harness, reasoning effort, provider, and version can materially change results.
PRIMARY METHODOLOGY
OPEN ORIGINAL METHODOLOGY ↗Read the evaluation before the rank.
Scores on this board are not interchangeable with scores from another test.