Small open models (30B parameters or fewer) to reach the capability level for end-to-end self-replication by the end of 2026
“If small models improve at their historical rate, they’ll match this capability level by the end of this year. Small weight-based replicators may be possible soon after.”
Context: Anthropic research post "AI models may be capable of self-replicating without access to a datacenter" (Frontier Red Team, 12 Oct 2026; Al Jazeera names the researchers as William Zhou and Daniel Freeman). "This capability level" is that of GLM-5.2, which completed end-to-end replication in Anthropic's simplified evaluations. The post fits open-weight models of 30B total parameters or fewer against Epoch AI's Capabilities Index (ECI), a gain of 13.7 ECI per year, and says "the trend line puts a small model at that level by the end of 2026", calling this a rough estimate. Reported by Al Jazeera and Unite.AI.
How we judge it
Our interpretation: correct if by 31 Dec 2026 an open-weight model with 30 billion or fewer total parameters, released on or before that date, scores at or above GLM-5.2 on Epoch AI's Capabilities Index (the measure Anthropic's estimate uses), per Epoch AI's published data, or Anthropic or another credible evaluator reports such a model completing end-to-end self-replication in an evaluation comparable to Anthropic's; wrong otherwise. A replicator spreading in the wild is not required. Unresolvable if no ECI data for qualifying models is published and no comparable evaluation exists.
Resolution criteria are our interpretation of the claim, written before grading. Methodology
- Said
- Resolve by
- Due in 55 days
- Verdict
- Open
- Stated probability
- None stated
- Specificity
- 3 of 3 (precise and measurable)
Share this scorecard
Post on X LinkedIn Share image (PNG)
Embed / Cite
The embedded card is live: it shows the current verdict, with the date the data was read. If a verdict is corrected later, the card changes too, so cite the page and its date for a fixed record.