news

Grok Imagine Video 1.5 Lite enters the Artificial Analysis board at Elo 993, with a rank range wider than the gap being claimed

The leaderboard prints rank 17 and a range of 16 to 18 for xAI's distilled video model, and the $0.02 per second headline is seven times cheaper than the tier the board priced.

By Priya Shenoy · 11 October 2026

Grok Imagine Video 1.5 Lite, xAI's cheaper cut of its flagship video generator, has landed on the Artificial Analysis text-to-video board at rank 17 with an Elo of 993, and the single most useful number next to it is the one the launch coverage skips: the board also prints a rank range of 16 to 18 for the model. The widely repeated framing is that Lite sits two places above Google's Veo 3.1, which is a gap narrower than the uncertainty the leaderboard itself attaches to the position.

What the leaderboard actually prints

On AA-Video-T2V v2.0, which ranks models on generating video with audio from a text prompt using Elo from pairwise human preference votes, Lite shows an Elo of 993 with a 95% interval of 983 to 1003 across 5,715 samples, a listed release of August 2026, and API pricing of $8.40 per minute. Its parent, Grok Imagine Video 1.5, sits at rank 11 with a range of 11 to 12, an Elo of 1045 (interval 1036 to 1054), 4,056 samples and $15.00 per minute. The two intervals do not overlap, so the 52-point gap between parent and child is real on this board, whatever the ordering against neighbours at rank 15 to 19 turns out to be.

The sample count is worth holding onto. 5,715 blind comparisons on one model is an order of magnitude more than some public video boards carry in total, including the one we looked at ranking 12 models on 1,394 votes. Depth of voting is the difference between a rank you can argue with and a rank you can only read.

Two prices for the same model, seven times apart

The distribution listing describes Lite as a faster, lower-cost model distilled from Grok Imagine Video 1.5, supporting text-to-video and image-to-video, and prices it "from $0.02 per second". The detail underneath that headline: $0.02 per second is the 480p rate, and the price rises to $0.14 per second at 1080p, with image input billed at $0.01 per image. $0.14 per second is $8.40 per minute, which is exactly the figure Artificial Analysis lists. The benchmarked configuration is the expensive end of the card, not the advertised one.

The same card also states that 1080p output is rendered at 720p and upscaled. So the tier that carries the seven-times price premium over 480p is not a native 1080p render, which matters for anyone buying Lite as a finishing model rather than a drafting one.

Artificial Analysis is not internally consistent either. Its image-to-video board lists Grok Imagine Video 1.5 at $8.40 per minute, while the text-to-video board lists the same model at $15.00 per minute, presumably reflecting different resolution or audio assumptions, though neither board states which. Lite itself appears on the text-to-video board rather than the image-to-video one, despite image-to-video being a documented capability on its xAI model page.

Nobody agrees on when it shipped

The release date is unsettled in a way that is unusual for a frontier lab. The Artificial Analysis row says August 2026. The OpenRouter listing says October 6, 2026. Model catalogue maintainers recorded a first sighting in xAI's developer docs on October 1 and 2, and noted that xAI publishes no release date for Lite and gives it no dated alias of the kind the full 1.5 carries. For a model whose entire pitch is price and speed against a dated competitor, a two-month spread on its own birthday is a reminder that the only figures worth quoting are the ones on a page with a name on it.

Sources: Artificial Analysis, OpenRouter, xAI Developer Docs.

Priya Shenoy: Tracks what AI video actually costs across the platforms that resell access to the same handful of models. Treats a pricing page as a claim, not a fact, until someone checks it.