Wan 3.0 Explained: Alibaba's 30-Second AI Video Model

Key Takeaways

• Alibaba Cloud opened public beta for Wan 3.0 on August 6, 2026. The headline capability is 30 seconds of video in a single generation pass, which makes continuous camera movement and one-take shot language possible without stitching clips.
• The genuinely new input is documents. Alongside text, image, audio, and video, Wan 3.0 accepts doc, xls, ppt, pdf, and md files plus web pages as creative references — turning a slide deck or spreadsheet into video.
• API access is not fully open yet and there are no published weights. Wan 3.0 is not callable on Siray today, though support is planned; the Wan model live on Siray now is Wan 2.6, in text-to-video and image-to-video.

What Alibaba Shipped

Alibaba Cloud announced the beta on August 6, 2026, confirmed the same day by Alibaba Group's official channels. It is positioned as an all-in-one generation and editing model from Tongyi Lab. What Alibaba has actually stated:

• 30-second single-pass generation. One call produces up to 30 seconds, not short clips joined afterward — which is what enables continuous camera movement and one-take shot language.
• Unified workflow. Wan 2.7 split text-to-video, image-to-video, reference generation, and editing across separate models. Wan 3.0 folds reference, editing, replication, and motion driving into one.
• Intelligent duration. The model proposes a clip length from your prompt instead of making you pick blind.
• Video extension. A separate tool extends an existing generation.
• Consistency. Alibaba emphasizes director-level camera work and stable characters, props, and scenes.

Alibaba is candid about the gaps: audio quality and typographic accuracy both still have room to improve.

The Actually New Part: Documents In, Video Out

Most video model launches add resolution, seconds, or reference slots. Wan 3.0's differentiator is what it will read.

Beyond text, image, audio, and video, it accepts doc, xls, ppt, pdf, and md files, plus web pages as creative references. Alibaba's pitch for this is "万物皆可生视频" — roughly, anything can become video.

The practical read: a quarterly deck becomes an explainer, a spreadsheet becomes an animated data segment, a product page becomes an ad. Alibaba calls out charts, data, and interface elements as areas it improved aesthetically, which fits that use case. Whether the output clears the bar to skip a human editor is unmeasured.

What It Costs

Alibaba Cloud's published API rates, per second of video:

The USD column is a conversion, not an Alibaba-published figure. Third-party aggregator listings for wan3.0-video disagree with each other and with these numbers, so treat the RMB rates as the reference and check Model Studio before budgeting.Note what the list implies: three tiers, topping out at 1080P. No 4K tier — the simplest available answer to the 4K rumors.

What Nobody Has Confirmed

Wan 3.0 launched without a technical report, model card, architecture disclosure, or parameter count. Several claims circulating now are unsupported:

• Native 4K. Not in the pricing tiers, not stated officially.
• Frame rate. No published fps figure. Any specific number you see is invented.
• Reference input limits. Described as omni-modal, with no stated cap on images, audio, or video accepted.
• Apache 2.0 release with 1.3B and 14B variants. Traces to a single social post, and that parameter pairing actually describes Wan 2.1 from February 2025.
• Benchmark rankings. Wan 3.0 is absent from the Artificial Analysis Video Arena, so "best video model" claims have nothing behind them.

On weights, Wan 3.0 is API-only. Wan 2.2 from July 2025 remains the series' last open-weight flagship. Alibaba did ship WanSong, Wan-Dancer, and Wan-Streamer under Apache 2.0 in July 2026, but those are adjacent models.Same pattern as Qwen Image 3.0: capabilities described, benchmarks absent.

What You Can Use on Siray Today

Wan 3.0 is not callable on Siray yet — its API is still rolling out on Alibaba's side. Support is planned. What you can call today:

On the 30-second question specifically: if long clips are what you came for, MiniMax H3 is live on Siray and reaches roughly 30 seconds via its Extend tool. That is stitched rather than single-pass, so it is not equivalent to Wan 3.0's claim, but it is available today.

All Siray endpoints sit behind one API key with normalized request and response shapes, so moving between models is a name change rather than an integration. Wan 3.0 is on the roadmap: once Alibaba opens the API, switching to it is the same one-line change on your side, with no new key and no new request format to learn.

FAQ

Is Wan 3.0 available through a public API right now?

Not fully. It entered public beta on August 6, 2026 and is reachable through Alibaba's own surfaces — Model Studio, the Wan site, Qwen's PC creation tool — but Alibaba described full API access as opening "soon" without committing to a date.

Is Wan 3.0 open-weight like earlier Wan models?

No. Wan 3.0 is API-only, with no weights on Hugging Face, GitHub, or ModelScope. Wan 2.2 from July 2025 is the series' last open-weight flagship.

Which Wan models can I use on Siray today?

Wan 2.6 for text-to-video and image-to-video. Wan 3.0 is not callable yet; Siray support is planned for when Alibaba opens the API.

Announced Now, Usable Later

Wan 3.0 is a real launch with a real price list, and document-to-video is a new capability rather than a spec bump. It is also in beta, closed-weight, unmeasured, and not fully API-accessible — so the honest position today is interest, not migration.

Create your free Siray account and start generating with Wan 2.6 now, with one API key across 300+ models — and be ready the day Wan 3.0 lands.