Transparency notice
Research claims
We distinguish a measured result from what that result might mean. This page supplies context for the research and competition statements on our company site.
Effective August 18, 2026
The reported ARC‑AGI‑3 result
This statement describes a result displayed on a public competition leaderboard. Public leaderboard values can change as submissions, evaluation infrastructure, competition rules, and organizers’ processes evolve. It does not state a final private score, placement, medal, award, replication, peer review, or performance outside the competition.
Sources · verified August 18, 2026: Kaggle competition and ARC Prize competition context.
What the metric does not establish
A single benchmark result is not evidence that a system has achieved artificial general intelligence. It does not, by itself, establish broad reliability, autonomy, safety, scientific validity, product readiness, or performance in real-world settings.
“Toward general AI” describes a research direction: studying capabilities such as exploration, world modeling, goal acquisition, memory, and planning in unfamiliar interactive environments. It is not a certification or claim that general intelligence has been reached.
How we aim to communicate evidence
- Name the object measured. We identify a public competition score as a public competition score.
- Separate observation from interpretation. A metric and our hypothesis about its meaning are different statements.
- State material limits. We avoid silently generalizing from one environment, dataset, or evaluator.
- Correct changing claims. If a public statement becomes inaccurate, we aim to revise or qualify it.
- Avoid implied validation. Participation, submission, ranking, or citation does not imply organizer or third-party endorsement.
Research areas
Our public description covers broad areas rather than specific unpublished methods: autonomous and multi-turn agents; efficient model compression and routing; multimodal and video reasoning; biological foundation models and AI for science; and interactive environments for studying behavior.
Descriptions of planned, experimental, or prototype work are forward-looking and may not become released products. Product pages should be consulted for current availability.
Photon47 competition submissions
Photon47 was submitted for consideration to the 2026 GWB Game Awards and Taipei Game Show’s Indie Game Award 2027. Submission does not mean the game was selected, nominated, shortlisted, awarded, sponsored, affiliated with, or endorsed by either program or its organizers. Sources verified August 18, 2026.
Corrections and questions
If you believe a research, benchmark, or competition statement is unclear or inaccurate, email [email protected] with the page, statement, and supporting information. We will review substantive correction requests.