Grok Imagine Video 1.5 Lite Debuts at #17 on the Artificial Analysis Video Board, 52 Elo Behind Its Parent for 44% Less

Two numbers make the case for G​rok Imagine Video 1.5 Lite, and neither is the one the launch note leads with. On Artificial Analysis's AA-Video-T2V v2.0 board the new G​rok Imagine Video 1.5 Lite sits at #17, Elo 993 (±10, 5,715 samples), listed at $8.40 per minute at 1080p default settings. Its parent, G​rok Imagine Video 1.5, sits seven rows higher at #11 — Elo 1045, $15.00 per minute. That is 52 Elo for $6.60 a minute, a trade that is unusual in video generation, where the tier below a flagship normally costs roughly what the flagship costs and simply does less. G​rok Imagine Video 1.5 Lite is not that. It is a distilled model with a real discount attached to it, and the discount is the story.

What landed, and when

G​rok Imagine Video 1.5 Lite is S​paceXAI's cheaper, faster cut of G​rok Imagine Video 1.5, and it is a distinct model with its own listing rather than a cheaper serving tier of the same weights. The interface has not changed: it takes text-to-video and image-to-video requests, 1-to-15-second durations in whole seconds, the seven aspect ratios the family supports (16:9, 9:16, 1:1, 4:3, 3:4, 3:2, 2:3), a first-frame image and additional reference images, and a callback URL for asynchronous completion. Generation is a two-step protocol — POST a request, get a job id and a polling URL back with a 202, then poll until the status flips to done and pull the bytes from a content endpoint. Failed generations are not billed.

The timing matters for how you read the ranking. Artificial Analysis dates the model to August 2026; the board entry itself is newer, and the announcement that put it on the board is what makes this a story this week rather than a footnote from the summer. A model can ship quietly and become news later, when the independent board finally has enough blind comparisons to place it — 5,715 samples in this case, which is a substantial panel for a video board.

Two boards, one of them verified

The announcement states that G​rok Imagine Video 1.5 Lite lands 17th on both the with-audio and the no-audio variants of the AA-Video-T2V v2.0 board, and that it takes a position on the board's quality-versus-price frontier. Of those claims we could confirm the with-audio row directly from the published leaderboard: rank 17, Elo 993, a 95% interval of 983 to 1003, price $8.40 per minute, credited to S​paceXAI. The no-audio placement and the frontier framing come from the announcement itself and we have not reproduced them independently.

It is worth being precise about what "Pareto frontier" buys the reader here, because the phrase gets thrown around. On a quality-versus-price plot, a model is on the frontier when nothing else is both better and cheaper. G​rok Imagine Video 1.5 Lite at Elo 993 and $8.40 a minute is competing in a band of the board where several models cluster: PixVerse V6 sits one row below at Elo 985 for $6.90, Kling 3.0 1080p (Pro) sits one row above at Elo 1000 for $10.08. The interesting claim is not that Lite beats any single one of them but that at this price it is close enough to the models above it to be worth the seat.

Where the discount came from

Three things are cheaper about Lite, and only one of them shows up on the price sheet.

The first is the number that the listing prints outright: 1080p output is rendered at 720p and upscaled. That is the single most useful sentence on the model's own description, and it is the one a buyer skimming a price table will miss. A Lite clip requested at 1080p is a 720p render carried up to 1080p dimensions. For social delivery that is often invisible. For anything that will be graded, cropped, or keyed, it is the difference between footage and an approximation of footage, and it is the reason the per-minute price can sit at 56% of the parent's.

The second is distillation. Lite is described as quicker and cheaper video generation distilled from G​rok Imagine Video 1.5 — the same relationship a small language model has to the large one it was trained to imitate. Distillation buys latency and cost and gives back a little quality, which is exactly what the board shows: 52 Elo of it.

The third is what the board cannot see. A distilled video model is not just less accurate on the aggregate preference vote; it is usually less reliable at the edges — long prompts, unusual camera moves, the reference-image conditioning that the full version supports more broadly. Blind pairwise voting averages those failures into a single Elo, and a single Elo is a poor predictor of whether the specific shot you need will come out.

What the numbers do not establish

The 993 is a preference score, not a fidelity score. The panel voted between clips without knowing which model made them, which makes it a strong measure of which clip people prefer and a weak measure of whether a clip matches a brief. A model can win votes with prettier color and lose the shot you asked for.

Price, meanwhile, is quoted per minute at a stated resolution and default settings. Change either and the comparison moves: drop to a lower tier and the per-second rate falls, extend a clip and the total rises. Anyone budgeting from a single per-minute figure is budgeting from a default, not from their own workload.

And there is no third-party evaluation of G​rok Imagine Video 1.5 Lite outside the arena — no published prompt suite, no reproducible quality harness, no independent latency benchmark. The Elo is the independent number. Everything else about the model, including its resolution behavior, comes from the vendor.

Who should actually switch

The decision is narrower than the launch framing suggests. If your output is short social video at 1080p or below, and your budget is the binding constraint, Lite at $8.40 a minute against the full model's $15.00 is close to a free upgrade — you keep the family's interface, its durations, and its reference-image inputs, and you give up a few percent of a preference score most viewers will never notice.

If your output ends in a grade, a crop, a composite, or a delivery spec, the upscaled-1080p detail is disqualifying, and the full G​rok Imagine Video 1.5 remains the model in this family you want.

If your question is which of the two to try first, that is a failover question rather than a commitment question. Neither model is served by OrcaRouter — we host neither G​rok Imagine Video 1.5 nor G​rok Imagine Video 1.5 Lite, and both are reached through the vendor's own API and several third-party platforms. What a router does cover is the layer around them: the prompt rewriting, the shot-list generation, the caption and metadata pass, and the model calls that decide which clip to keep. That layer runs on OrcaRouter as one API across 200+ models at provider list price with 0% markup, with automatic failover across providers and a routing DSL that composes several models into a single call. A per-second video price never touches that bill, which is precisely why the text side is where a routing decision pays for itself.

What to watch next

Two things would settle the open questions. The first is the no-audio board placement, which would confirm whether Lite's ranking holds when the audio channel is removed — if the gap to the parent widens sharply there, the audio track is carrying more of Lite's Elo than its visuals are. The second is any independent evaluation of the upscaled 1080p output, since the whole value proposition of a $8.40-a-minute video model rests on whether a 720p render carried to 1080p survives the delivery pipeline it is being sold into.

Until then the honest summary is a narrow one: G​rok Imagine Video 1.5 Lite is the cheapest way into the G​rok Imagine Video family that anyone has measured, it gives up 52 Elo to do it, and it gives up native 1080p detail as well — a trade that most social-first teams should take and that anyone finishing footage should not.