The date arrived. The weights did not. — a vendor-promise ledger for model release dates
PromiseLedger is a planning aid, not procurement advice; verify each promise against the vendor's own channels. 1. The receipt StepFun put a date on its own calendar: Step 5's BF16 open weights would ship on October 15. As of 01:55Z that day, the expected HuggingFace repo was a 404. A same-day re-check returned an auth error — the repo was gated or unfilled, not weights. The vendor stayed silent.…
StepFun promised that the BF16 open weights for model Step 5 would ship on October 15. However, as of 01:55 UTC on that day, the expected HuggingFace repository was completely inaccessible, indicating the weights were either gated or not yet available. The vendor did not provide any explanation.
Lab-announced model-availability dates, as reported by forkast.news, are often the most unpredictable and costly for enterprise deployments when such deadlines are missed. These dates are specific and announced by the lab, but now proven to be highly volatile to plan around.
When planning production around unreleased model weights, it is crucial not to rely on them. Instead, create an evaluation set and serving path using a model that can actually be downloaded. This way, when the new weights are released, they can be swapped in without issues. The evaluation set should be portable, and the serving path should be real and tested.
The promise ledger provides a structured way to track dates your roadmap depends on. Each date is accompanied by evidence, a status engine, and a risk score ranging from 0 to 100. If a promise is missed with no contingency plan, the risk score caps at 49.
For Step 5, the vendor initially announced a firm date. However, when the HuggingFace repository went offline, the promise was missed, and without an official contingency plan, the risk score was hard-capped at 49. Another model, Beam (by Reflection), has a vague announcement of "later this month" with no specific date. Vague promises like these score down because they cannot be calendarized or planned for.
The third model, Kolibri (by Aleph Alpha), has already shipped on October 3, representing a controlled case with zero date risk.
The PromiseLedger tool was developed to address these issues. It allows users to track model release dates critical to their infrastructure, classify the promise tier, and apply the slip playbook (which includes fallback options, ensuring evaluations are portable, and setting up alerts for any changes in the promised dates). Priced at a one-time fee of $39, PromiseLedger is offered by Haku at https://vittoriali.gumroad.com/l/ncsrl.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
