AI video releases and shutdowns in 2026: one dated timeline

Between January and October 2026 the vendor pages we read list at least twelve dated launches, deprecations and shutdowns across video, voice and ad tooling. The table below puts them in one order so you can see how often a model id you depend on changes under you.

What the sequence shows

Three patterns are visible without any outside data.

• Google moved a video model from preview to GA in 58 days, and announced two shutdowns with only 15 and 63 days between notice and cutoff.
• Luma shipped model or workflow releases in January, June, August and October.
• A shutdown can arrive with no direct replacement. OpenAI's guide states the Sora API ended on September 24 with no direct replacement API.

How to use a timeline like this

Keep your own copy with one extra column: which of your jobs call each id. Then a deprecation notice becomes a query instead of a search. On Sume, model ids are bare catalog ids such as seedance-2.5 or gemini-omni-flash-1.1, and each poll response reports the model that ran, so a log of job id plus model id is enough to find affected outputs.

Before you adopt anything from the table, read the vendor page again. Dates and limits on launch-week pages change, and this list reflects what the pages showed on October 3, 2026. The catalog endpoint documented in Video generation lists what is callable on Sume today.

Reading the dates against your own release plan

A useful exercise is to lay your own roadmap over the table. If you shipped a feature in June on a Google preview, the deprecation notice of June 15 and the Omni preview of June 30 landed inside the same month. If you built on Sora, the Sora API cutoff of September 24 fell four days before ElevenLabs introduced Eleven v4, which is the sort of overlap that stretches a small team.

Cadence also matters for budgeting engineering time. Four Luma releases in ten months, three Google video or voice milestones in twelve weeks, and two shutdowns announced together mean a fixed, recurring cost of re-testing. Plan for it the way you plan for dependency upgrades: a calendar slot, an owner and a standing smoke test.

None of this ranks the vendors. A vendor that ships often gives you more options and more churn at once, and a vendor that ships rarely gives you stability and fewer options. The table only records what happened.

Gaps in this list

The table is not exhaustive. It covers the pages we read for this batch, and it omits vendors we did not check. Treat an absent vendor as unchecked, not as quiet.