AI video model news, early October 2026: what to re-test
Kling 4.0 early access, Kling on Runway MCP, the Sora API shutdown and Gemini Omni Flash 1.1, with dates, sources and the one check each change calls for.
Four things changed for AI video builders in the last two weeks: Kling opened Kling 4.0 early access on 2026-09-28, Runway put four Kling models on its MCP on 2026-09-18, the Sora API shut down on 2026-09-24, and Gemini Omni Flash 1.1 has been generally available since 2026-08-27. Of these, only Gemini Omni Flash 1.1 has a matching row in Sume's catalog today; Kling 4.0 is not listed (Kling 3 is), and Sora and Veo are not in the Sume video catalog.
Every date below was read on 2026-10-02 from the page named in the table. The Veo item is different: it comes from a search-result summary of a PixVerse article that I did not open, so it is labeled as reported.
What happened, and where is it written?
Each row names its source and whether it was read on the page itself.
| Date | Change | Source |
|---|---|---|
| 2026-09-28 | Kling 4.0 early access begins; full launch due October 2026 | Kling blog (read on page) |
| 2026-09-24 | Sora API discontinued (app closes 2026-04-26) | The Decoder (read on page) |
| 2026-09-18 | Four Kling models added to Runway MCP for paid plans | Runway changelog (read on page) |
| 2026-08-27 | Gemini Omni 1.1 Flash generally available | geotoolbox.ai (read on page) |
| As of 2026-09-20 | Veo 3.1 reported as the production Veo; no official Veo 4 date, id or price page | PixVerse summary (reported, search result only) |
What does each change mean for an integration?
One check per change keeps this from turning into a research project:
- Kling 4.0: nothing to call yet. Keep
kling-3working and watchGET /v1/videos/modelsfor a new Kling id. - Runway MCP Kling models: if your agent uses Runway MCP, confirm which Kling names it can reach on your plan. Sume lists
kling-3and the Kling 3.0 Motion Control route. - Sora: if any code path still names a Sora model, remove it and decide the replacement. Sume lists no Sora model.
- Gemini Omni Flash 1.1: it is
gemini-omni-flash-1.1on Sume, 3 to 10 seconds, 360p to 4K, 16:9 or 9:16, always with audio. - Veo: do not plan around a Veo 4 date that no Google page has given.
Which of these can you call on Sume today?
Only the Omni row has a Sume id matching the news. For the others, the catalog shows what is closest, and GET /v1/videos/models shows what is current when you read this.
| Trend | Sume id today | Limits |
|---|---|---|
| Kling 4.0 | None listed | Not callable |
| Kling (3.x) | kling-3 | 4 to 15 s, 720p or 1080p, 16:9, 9:16, 1:1 |
| Gemini Omni Flash 1.1 | gemini-omni-flash-1.1 | 3 to 10 s, 360p to 4K, 16:9 or 9:16 |
| Sora | None listed | Not callable |
| Veo 3.1 | None listed | Not callable |
How do you keep up without re-reading the news?
Poll the catalog on a schedule, keep model ids in config, and keep a small set of prompts whose outputs you re-run when a row changes. The news tells you when to look; the catalog response tells you what you can send.
Sources
Related posts
More in Models
- POST /v1/avatar-1.0/fabric is test-only: use veed/fabric-1.0 instead
The experimental avatar-1.0/fabric route is a temporary comparison endpoint that may be removed. How it differs from image-to-video and what to build on.
- Best AI video model for ads in 2026: check limits, not only Elo
A leaderboard cannot tell you if a model fits your ad length. Sume's docs list per-model limits: 3 to 30 seconds, ratios, and 480p to 4K.
- ByteDance MPA copyright pact: what Seedance users should check
ByteDance signed a copyright agreement with the MPA on Aug 17, 2026 covering Seedance. What was disclosed, what was not, and how to handle a refused request.
- Can GPT-6 Sol, Luna or Astra generate images or video?
No. OpenAI lists GPT-6 Astra, Sol and Luna as text-output models with image input. Where image and video generation live instead, and how Sume splits it.
Written by Sume