Aiqrion v0.6 is in alpha — model output is not production-grade and may be incorrect. Aiqrion does not claim to beat Claude / ChatGPT / Gemini / Grok yet.

Aiqrion training datasets

Aiqrion never trains on a dataset that hasn't passed license / PII / safety / contamination review. v0.6 ships the review for the candidate pool; only train_allowedrows are eligible for training mixes.

Loading dataset decisions…

v0.6 training mix policy

Aiqrion v0.6 training mix is intentionally tiny: aiqrion-v0.4-curated only. Closed-model distillation datasets (Claude / Opus), non-commercial datasets (AtomBlock-WebUI), benchmark held-out splits (SWE-bench / GSM8K test / MMLU-Pro / ParseBench), and gated datasets (HLE) are explicitly excluded. See docs/datasets/v0.6-training-selection.md.