Vidu Q4 Preview 對決 Grok Imagine Video：一月還在 Arena 領先的模型，如今位居第九

Three numbers describe this matchup, and none of them are a spec sheet. On the Artificial Analysis image-to-video board, Vidu Q4 Preview sits third at 1,179 with 5,543 votes behind it. Gr​ok Imagine Video 1.5 sits ninth at 1,098. The January 2026 build of Gr​ok Imagine Video, still listed and still callable under its own name, sits fourteenth at 1,072 — and the two Gr​ok entries are priced at $8.40 and $4.20 per minute respectively, which means the newer one costs exactly twice as much for twenty-six more Elo.

Vidu Q4 Preview is one build, released 7 October 2026 by Shengshu Technology as a first public preview ahead of the full Q4 model, in two modes: Image-to-Video and Reference-to-Video. Gr​ok Imagine Video is a family with at least three entries across two boards. The comparison is really about what happens to a leaderboard position over nine months, and the Vidu release is the thing that arrived to make the point.

同一模型的兩個建置，Elo 相差二十六

Artificial Analysis，讀取於 2026年10月8日，保留音訊，圖像轉影片排行榜（AA-Video-I2V v1.0）：

• Vidu Q4 Preview — 第 3 名、1,179 ±10、5,543 個樣本、2026 年 10 月、每分鐘 $7.20

• Gr​ok Imagine Video 1.5 — 9th, 1,098 ±8, 5,267 samples, May 2026, $8.40/min

• grok-imagine-video — 第14名、1,072 ±7、14,371 個樣本、2026年1月、每分鐘 $4.20

而在文字轉影片（AA-Video-T2V v2.0）方面，則另設一個獨立排行榜，採用其自身的評分尺度：

• Gr​ok Imagine Video 1.5 — 11th, 1,045 ±9, 4,056 samples, May 2026, $15.00/min

• Gr​ok Imagine Video 1.5 Lite — 17th-18th, 993 ±10, 5,715 samples, Aug 2026, $8.40/min

• Vidu Q4 Preview — 不存在

Two things jump out of the first block. The January build of Gr​ok Imagine Video carries the most votes of any model in this comparison — 14,371, nearly three times Vidu Q4 Preview's 5,543 and nearly three times the 1.5 build's. When a model launched as the arena leader in January, that is what the sample count looks like in October: the community voted on it extensively, and then stopped. The score is stable and well-measured and no longer near the top.

The second thing is the price inversion. Gr​ok Imagine Video 1.5 is twice the per-minute rate of its January predecessor and twenty-six Elo ahead of it. Twenty-six Elo is just outside the two intervals combined — ±8 and ±7 — so the improvement is probably real, but it is the kind of margin that comes from a generational refresh rather than a new architecture, and it is being sold at double. Meanwhile Vidu Q4 Preview is eighty-one Elo above the 1.5 build and priced below it, at $7.20 against $8.40. That is a fresher model beating an older one on both axes, which is what is supposed to happen and rarely does this cleanly.

The third oddity sits between the boards. The same Gr​ok Imagine Video 1.5 that is $8.40 per minute on image-to-video is $15.00 per minute on text-to-video, where it also scores lower by the other board's scale. If you are planning a Gr​ok workload, the input modality you choose changes the rate by roughly eighty per cent.

Grok Imagine Video 實際上能為你帶來什麼

這款模型並不差，本文也不會假裝它差。有兩件事它做得到，而 Vidu Q4 Preview 做不到：

It takes text prompts. Vidu Q4 Preview's product page lists exactly two modes — Image-to-Video and Reference-to-Video — and its API documents /ent/v2/img2video and /ent/v2/reference2video and nothing else. There is no text-to-video endpoint. Gr​ok Imagine Video 1.5 and its Lite sibling both appear on the text-to-video board, so if your pipeline starts from a written prompt, one of these two models is available to you and the other simply is not.

It has a cheap tier. On the image-to-video board the January build is $4.20 per minute against Vidu Q4 Preview's $7.20 on that same board; the Lite tier's $8.40 sits on text-to-video, a separate board with its own scale. If you are generating high volume and can accept a fifth-place-tier score, the low end of the Gr​ok family is the cheapest thing in this article — the January build at $4.20 carries 1,072 Elo, which is 107 below Vidu Q4 Preview but also $3.00 per minute cheaper.

Vidu Q4 Preview 能為你帶來什麼

其中的 Vidu 這一方，在雙方共有的軸線上，是更強大的生成模型，而且擁有龐大得多的參考預算。

• Resolution — Vidu Q4 Preview up to 4K at 10-bit colour (2K and 4K) vs Gr​ok Imagine Video not documented at 4K on these boards

• Clip length — Vidu Q4 Preview 3-16s for Image-to-Video and 1-16s for Reference-to-Video vs Gr​ok Imagine Video not published to the same detail

• Reference inputs — Vidu Q4 Preview up to 15 reference images and up to 3 reference audio clips vs not published to that level for Gr​ok Imagine Video

• 音訊預設 — Vidu Q4 Preview 的圖片轉影片端點帶有一個 audio 參數，預設為 true，因此除非你將它停用，否則聲音會隨畫面一同出現

• 文字轉影片——缺席，而這是這份清單上唯一朝相反方向走的一項

Every Vidu figure above is the vendor's own and should be read as such. The one number in this article that is independently measured is the Elo, and on that Vidu Q4 Preview is eighty-one points clear of Gr​ok Imagine Video 1.5 with a thousand fewer votes. Eighty-one Elo is roughly five times the combined interval width. It is a real gap today and it is dated October 2026, which means it is a gap about a different moment than the Gr​ok scores sit in.

預覽版本是公平的對手嗎？

生數科技（Shengshu）的說法異常直白：Vidu Q4 Preview 搶在完整的 Q4 模型之前推出，好讓創作者能在真實專案中使用，並協助形塑最終正式版本。其定價被形容為促銷價。最終定價、支援的解析度、功能可用性與使用條款可能因方案與地區而異。API 文件裡夾帶了一個拼錯的別名，viduq4-previerw，就緊挨在正確的 viduq4-preview一旁，而這種細節，正式出貨的產品通常會清掉，預覽版通常不會。

所以，這 81 Elo 是由一個其自家供應商表示並非最終版的版本所拿下，而且是在其自家供應商稱為促銷價的價格下取得。這兩點都對 Vidu 有利——模型很可能會變得更好，價格也很可能會上漲——但這也意味著這兩個數字都不是承諾。任何想依據本頁面編列預算的人，都應該預期費率會在 Q4 發布時變動，而且屆時應該重新細讀榜單，而不是只看排名。

If the preview build turning over every few weeks is the problem, that is exactly the case automatic failover exists for: one OpenAI-compatible endpoint across 200-plus models, with provider list prices passed through and no markup added, so a promotional rate ending is a line item change rather than a renegotiation, and a route going unavailable falls through instead of failing the request. Being straight about coverage — Vidu Q4 Preview is not an OrcaRouter route, and neither is Gr​ok Imagine Video. The video models we do serve, including MiniMax H3, sit on that same key and endpoint as the text models.

狹義的答案

If you want the best image-to-video output and your input is a frame, Vidu Q4 Preview is eighty-one Elo ahead of Gr​ok Imagine Video's current build at a lower measured rate, and there is no strong argument for the Gr​ok side beyond price floor. If you want the cheapest thing that scores respectably, Gr​ok Imagine Video's January build at $4.20 with 1,072 Elo behind 14,371 votes is a well-measured, well-voted, deliberately cheap option and Vidu has nothing at that price point. If your input is a sentence, the Gr​ok family is the only one of the two families in the room.

What would change it: the full Vidu Q4 release, which could add modes or narrow the price gap; and AA-Video-I2V v2.0, announced as coming with a new taxonomy and a fresh vote pool, which would reset every number on this page rather than adjust it. Until either lands, the position worth remembering is the one the sample counts tell you — Vidu's 5,543 votes are one day old, Gr​ok's January build's 14,371 are nine months old, and a leaderboard is a photograph, not a standing.