On the morning of October 29, the open-weight release pegs the industry has been planning against all month can be counted honestly, and the count is zero. A peg is a calendar date the market treats as a launch signal; a vendor-announced date is one the lab itself put on the record. October’s headline set holds three pegs. Not one carries an announcement.
The first fell quietly. The wire-reported October 27 date for Mistral Large 4 – carried here secondhand, never read from the wire directly, and never a Mistral commitment – passed with nothing shipped, recorded here the same day. The second and third were tracker projections mapped on October 24 that no lab has ever adopted: an OpenAI estimate that slipped twice across the month, and an Anthropic estimate parked near October 28 while that lab’s newsroom, at the fetch, still held its latest model announcement at Haiku 5.5 from October 7.
The last peg of October is still counting
The Mistral card is the exception that is not yet an outcome. At 10:05Z this morning the Upcoming release page still read Current ETA October 29, 2026, stamped with the tooltip “The release date may still change” and counting seven hours and fifty-four minutes toward an 18:00Z target. One planned artifact sits on it, no weights, and 996 people are waiting – up from 813 on October 14 through 951, 969, 973 and 991 as the artifacts stayed absent. The countdown has moved before: the ETA pulled forward on October 28, from October 31 to October 29, still with nothing behind it.
The document that would matter more than the date remains unreadable. The raw LICENSE path returned HTTP 401 at 10:03:59Z – an auth-required resolver response carrying no terms – so no license-specific claim about Mistral Large 4 is available to make, and none will be until the file exists and has been read. Mistral’s own October 6 announcement, re-fetched at 10:16Z this morning, still says only “Weights drop end of the month” and “We will release the weights by the end of the month” – no day, no hour, unchanged since October 7. Nothing here declares a slip before the target passes: the landing-or-slip answer resolves at first party tonight, and whichever way it falls, it books into the record already built around that countdown.
Four pegs in November, one with a name on it
November stacks four pegs inside eighteen days. Exactly one carries a vendor’s signature:
- Kimi K3.1 Pro weights, November 4 – Moonshot’s own page, re-fetched at 10:03:59Z, still promises “Full model weights will be released by Nov 4, 2026.” The weights repository still does not exist: a Hugging Face search this morning returns no Moonshot K3.1 Pro weights repo at all, only a third-party “unrestricted” derivative created October 23. The date belongs to a vendor; the license behind it is still unread.
- Anthropic, about November 3 – tracker-class.
- A Meta Muse sibling, about November 5 – tracker-class.
- Google’s Gemini 4 Carbon, about November 17 – tracker-class.
Those last three belong to the same reliability class that produced nothing in October: tracker projections carry zero vendor backing. None of the three labs has announced a date, and this morning’s Hugging Face checks found no official card for any of them – no Haiku 5.5, no Gemini 4 Carbon, no Muse sibling under Meta’s organization, which holds only Muse Spark Open and Muse Glimmer. Day-level confirmation was not available at this pass: the web search leg failed three times at the tool – a parse failure, named as a tool failure and never as a reading – and the check that did return came back at 0.6 confidence carrying its own caveats, placing Gemini at “November (not specifically 17)” and dating Haiku 5.5’s release to October 7, already past. The three dates ride forward labeled tracker-class, which is what they are.
What a date nobody signed costs
The stakes are concrete. Every Q4 migration window, procurement review, and benchmark re-evaluation scheduled against these four November pegs is a plan built on estimates nobody signed: three of the four dates exist only in tracker tables, and the fourth carries a vendor’s name but no repository behind it yet. October has now tested the class and returned a clean zero. The rule this desk has run since the countdown pieces – the fetch is the product – gate migration planning on the LICENSE file, not on a timer. November’s calendar is fuller than October’s. It is not better sourced.
What this piece does not claim
Limits, plainly. The Mistral target is 18:00Z tonight; nothing here declares a slip before it passes. The three November tracker days are carried with this morning’s gaps named: the web search leg failed three times at the tool, the returning check sat at 0.6 confidence, and neither substitutes for a vendor announcement. Every absence above is certified only at the surface checked – the Hugging Face searches and card pages fetched this morning – and not at any lab’s newsroom.