28-model tracking table: half publish no training cutoff, Gemini 4 Carbon 8 months stale at launch
Joozio · reddit · 2026-11-16
A Reddit user hand-maintains a table of release dates and published training cutoffs for 28 models from 8 labs, each linked to the lab's own docs. Key findings:
- Only 13 of 28 models have a published cutoff. OpenAI, Anthropic, Google and xAI publish for every model; Mistral, Alibaba and DeepSeek publish none; Meta only for Llama 4, none for its five Muse models.
- Gemini 4 Carbon (Nov 10) and Gemini 4 Hydrogen (Nov 12) both list a March 2026 cutoff — roughly 8 months stale on launch day; Fable 5.2 and GPT-6.1 Terra are about 5 months behind.
- On the "search fixes it" argument: the author gave 16 models websearch plus 40 questions (2,000 calls, $6.53). GPT-6 Astra decided correctly 80/80 times, but 5 of 16 still claimed Harald V is king of Norway with the tool in the request — the same stale weights decide whether to search.
He notes the table's blind spot: for the 15 unpublished models, release date is only an upper bound, flattering exactly the labs that say nothing, and asks for ways to externally estimate cutoffs (e.g. probing for dated events).
More from Models
- Stealth Anthropic model spotted in Devin build as METR chart shows frontier pace up 1.4x — daniel_mac8 · 2026-11-16
- Eval Firm Trajectory Told 40 Agents "NEVER Stop" in Harness Without Telling Epoch — peterwildeford · 2026-11-16
- METR time-horizon chart updated: doubling every 1.7 months since Opus 4.5, 'Model 4' ~38h — ben_j_todd · 2026-11-16
- Gemini Pro user says its coding agent rejects every Three.js/SVG prompt as 'illegal' — Necessary-Buy-7410 · 2026-11-16
- Opus 5.5 caught using account email to identify user in supposedly memory-less session — birchlse · 2026-11-16
- Claim: Claude Fable isn't a tier above Opus — it's just rebranded Opus — iruletheworldmo · 2026-11-16