Internal-to-public deployment gap at the frontier
1What this measures
Months between a frontier model's internal availability and its public release, from the one documented case (Anthropic's Mythos class: internal in February, public in June 2026, per Ryan Greenblatt on the Dwarkesh Podcast, with Fortune's release report for the public leg).
Why it matters. AI 2027 turns on labs keeping their best models internal. A gap that grows is the scenario's signature; a gap of months is ordinary product testing.
- Proxy types
- model_release
- Unit
- months
- Cadence
- quarterly
- Valve
- invention to product
2How we track this
- series
dwarkesh.anthropic_mythos.internal_to_public_gap_months.pt - series
fortune.anthropic_mythos.public_release.pt - source Dwarkesh Podcast transcripts · default tier 6 · Substack; short quotation
- source Fortune · default tier 5 · Fortune Media; short quotation
- Normal band
- ≤ 6.00 months
- Fast band
- ≥ 12.0 months
- Falsifying
- —
Normal = a gap under six months, ordinary red-teaming and productisation; fast = a year or more, a lab running internally on a model the public does not have. Between is `emerging`.
Applied to dwarkesh.anthropic_mythos.internal_to_public_gap_months.pt.
3Tracker interpretation
Four months, on one hedged podcast statement. Enough to publish, not enough to lean on.
4Evidence
2 observations. Hollow points are disputed (see counterevidence). Every point links to its observation.
5Status and reasoning
Evaluator: 4 (Dwarkesh Podcast transcripts, 2026-06-30) is inside the consistent band (hi=6). Auto-reason; band rationale: Normal = a gap under six months, ordinary red-teaming and productisation; fast = a year or more, a lab running internally on a model the public does not have. Between is `emerging`.
6Timeline notes
- 2026-06-30 anthropic_mythos · 4.00 monthsas of 2026-06-30
7Counterevidence
What cuts against this reading
A single datum from a third party speaking from memory; internal availability is not internal deployment at scale; restricted partner access blurs the public date.
8Update history
- 2026-09-10unmeasured to consistent with normalconf — → 45 · evaluate
Evaluator: 4 (Dwarkesh Podcast transcripts, 2026-06-30) is inside the consistent band (hi=6). Auto-reason; band rationale: Normal = a gap under six months, ordinary red-teaming and productisation; fast = a year or more, a lab running internally on a model the public does not have. Between is `emerging`.
9Confidence
45 / 95 — limited or vague evidence
Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; below 50 limited or vague.