AI video generation model releases: the dossier's newest dated item and what it does not say
The newest dated development in this dossier is Ideogram 4.5, announced on 4 October 2026 as a multi-turn image editing model with four quality modes at native 2K, not a video model. The dossier contains no dated release of an AI video generation model, so that story cannot be sourced from it.
The dossier offers no dated release of an AI video generation model. The newest item in it that touches generative media at all is Ideogram 4.5, posted on 4 October 2026, and it is an image editing model. Anyone looking for a video model launch will have to look outside these 15 sources.
The rest of the dossier is equally clear on this point. There is a report about Google's Gemini subscription tiers, a paper on text entropy filtering, an OpenAI security disclosure, a Chinese space plane observation and a study of community archives. The only video in it is explainroo, an open source tool that renders narrated MP4s from a script, and it was not released in this window: its README lists version 0.4.1 on 20 October 2026 and 0.4 on 15 October, both of which appear to postdate the surrounding coverage, so those dates cannot be treated as reliable release news.
What the dossier does document
Ideogram 4.5 is the freshest product announcement available. The company's own page says the model is live in the Ideogram app, MCP and API, with Generate and Precise Edit endpoints and four quality modes. Pricing is fixed per image: $0.008 for a low-quality reference-based generate or an edit, $0.03 for low, $0.06 for medium, and $0.22 for high. The page claims the model eliminates artifact buildup across edits, which would make multi-turn editing viable, and it says images can be edited at high resolution without downscaling.
Those are vendor claims, not independent tests. The page also states that each image has a fixed price regardless of prompt length, reference count or output resolution, and that no subscription is required.
On the same day, Google's subscription changes drew the most attention. 9to5Google reported on 3 October that, per an updated support document, users without a Google AI plan will be limited to Flash-Lite from October 9, losing access to 3.6 Flash and 3.1 Pro, while $4.99 per month AI Plus subscribers will be limited to Flash-Lite and Flash. The Decoder's write-up adds that all three models require AI Pro at $19.99 per month or AI Ultra at $99.99 or $199.99, and frames the change as a mirror of OpenAI's freemium model. It also notes that Google may be preparing for Gemini 4 Argon, and that 9to5Google says Argon will reach AI Ultra subscribers first.
Neither outlet ties the tier change to video generation. The dossier gives no figures for Gemini's video output, and no Google statement about it.
Where the recent material actually sits
Most of the remaining sources are research or security. OpenAI's security page, dated 4 October, describes a coordinated campaign to extract protected reasoning from its models, with activity first seen in the first week of July, high-volume spikes on 24 and 25 July of 16,000 requests from more than 4,000 users, and full disruption by 28 July across a cluster of more than 15,000 users. OpenAI attributes a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi. It says operators did not break encryption or reach stored conversations, and that the technique is not unique to OpenAI's models.
Anthropic's voice data request, reported by BleepingComputer on 4 October, is a product and privacy change. The prompt reads, in the outlet's quotation, "Allow us to use your voice data to improve our AI models," and the setting is off by default, separate from chat training, and reversible from settings. No model release is involved.
The research side is thinner. Lewis Mitchell's arXiv paper, submitted 1 October, reports that a non-parametric Kontoyiannis entropy rate filter, computed from raw text with no model, beat logprob-based filtering in a six-generation QLoRA experiment on Llama-3.1-8B: $+42\%$ unique trigrams, $+30\%$ vocabulary and $-19\%$ repetition, all at $p < 0.001$, while the logprob baseline showed no significant benefit ($p > 0.23$). Jerry Gamblin's 4 October post runs Cloudflare's Clef models locally over 27,489 CVEs and finds that 989 of 23,736 non-kernel descriptions, 4.2%, state no security impact.
Two items are openly speculative. ResearchChamber.fun, posted 4 October, runs games in which a model has a pain pattern switched on, built on work by Tagliabue, Dung and Berg in 2026. Alex Zhang's blog post, also 4 October, argues that language model input and output shapes have barely changed since ChatGPT and asks whether architectures should be designed around the harness instead. Neither reports a product.
The releases that are dated
Three items in the dossier do have firm dates, and none is a video model. China's Shenlong space plane launched on 6 February 2026, and LeoLabs said it detected an unknown object near the craft at 02:30 UTC on 22 June, later catalogued by the U.S. Space Force according to Jonathan McDowell, as Space.com reported on 4 October. OpenAI's Dev Day, covered by TechCrunch on 29 September, centred on agentic assistants called Dots and on turning ChatGPT into an app surface, with Sign in with ChatGPT and 16 launch partners including Cognition's Devin, Notion, Vercel, T3, OpenClaw and Dactyl. The chatbot, TechCrunch says, has 1.2 billion weekly users, and none of the announcements concerns video.
Square Enix has a date, at least. Final Fantasy 7 Revelation arrives on 8 April 2027 for Switch 2, PC, PS5 and Xbox Series X/S, and director Naoki Hamaguchi told Radio Times, as reported by Eurogamer on 4 October, that the disc is a "key disc" holding only the opening. Sony, Eurogamer notes, said on 1 July that it will stop producing discs for PlayStation games from January 2028.
Mellon's archive study, released 4 October, surveyed 336 community archives. Its president, Elizabeth Alexander, is quoted saying they are "the strongholds of our American history." The foundation says it has given more than $40 million to over 100 such archives since 2020.
For video generation, the honest answer is that this dossier has nothing to report yet.
Sources: the-decoder.com, 9to5google.com, ideogram.ai, openai.com, bleepingcomputer.com, techcrunch.com, arxiv.org, jerrygamblin.com, researchchamber.fun, alexzhang13.github.io, space.com, eurogamer.net, mellon.org, github.com/vincentsch/explainroo, and the 2017 Berkeley technical report.
Sources
15- 01Ideogram 4.5: The most precise edit modelEN
- 02Gemini app limiting what models free and AI Plus users can accessEN
- 03Google's new Gemini tiers cut free users to its weakest model and lock $5/month subscribers out of ProEN
- 04Disrupting a coordinated model-distillation campaignEN
- 05Anthropic asks Claude users to share voice data for AI model trainingEN
- 06No Model Required: Text Entropy Rate Filtering Mitigates Iterative Fine-Tuning CollapseEN
- 07Decision Models vs. CVE DataEN
- 08What happens when an AI model is put in a "pain" state?EN
- 09Language Model "Shape"EN
- 10China's space plane appears to have released a mystery object in orbitEN
- 11OpenAI's latest features take direct aim at the app store modelEN
- 12FF7 Revelation director says he "did his best" to fight for a full disc releaseEN
- 13Mellon Foundation Releases the First National Study of Community-Based ArchivesEN
- 14Explainer videos and product demos made by your AI agentEN
- 15Threat modeling and circumvention of Internet censorship (2017)EN
All figures and quotations in this text come from the sources listed below.
Content prepared by the editorial team with AI assistance.
Comments
0- No comments yet — be the first.