Grok Imagine Video 1.5 Agent enters the Text-to-Video Arena at number five, putting SpaceXAI level with Wan 3.0 and FLUX 3 Video

The Text-to-Video Arena has a new entrant worth paying attention to. Grok Imagine Video 1.5 Agent from SpaceXAI has just landed at number five on the Artificial Analysis leaderboard with 1491 points, sitting a mere three points behind both Wan 3.0 and FLUX 3 Video, which each hold 1494. For AI filmmakers who have been tracking the competitive landscape, this is not a trivial result. It means a model that did not exist on this chart until now has arrived already trading blows with the current consensus picks at the mid-to-upper tier.

The placement also puts Grok Imagine Video 1.5 Agent ahead of Dreamina Seedance 2.5, Seedance 2.0, and MiniMax H3, which until recently were the names most creators reached for when they wanted reliable quality at scale. SpaceXAI has not been a loudly discussed presence in the AI video conversation, so a top-five debut in a human-preference arena makes this a moment to recalibrate assumptions about who the serious contenders are.

What the leaderboard position actually tells you

Arena scores are derived from human preference votes, not synthetic benchmarks, which makes them one of the more honest signals available. A three-point gap between Grok Imagine Video 1.5 Agent, Wan 3.0, and FLUX 3 Video is well within the noise of head-to-head voting. In practical terms, these three models are statistically tied at the time of writing. What that means for a working filmmaker is that SpaceXAI's entry is not a "close but not quite" result: it is a genuine peer to the models that have been driving most serious AI video production over the past several months.

What was not said

The Arena announcement from @arena gives the score and the ranking, but it does not tell you much about where the model excels or struggles. Arena scoring is aggregate: a model that is very good at one style and average at everything else can score similarly to one that is consistently solid across the board. Until independent creators start publishing detailed breakdowns, it is not possible to say whether Grok Imagine Video 1.5 Agent is a specialist or a generalist.

Pricing and access terms have not been announced alongside this result. It is not yet clear whether this model is available via API, through a consumer product, or on a waitlist. The "Agent" designation in the name suggests some form of agentic or multi-step generation pipeline, but no technical specifics have been published to accompany the Arena entry. That is a meaningful unknown for anyone planning a workflow around it.

There is also no information on output resolution, maximum clip length, or consistency controls, all of which matter more to a filmmaker than a headline score. Seedance 2.5 and Wan 3.0 have both been stress-tested by the community at this point, with known strengths and known failure modes. Grok Imagine Video 1.5 Agent has essentially no community track record yet.

How this sits against the current alternatives

Wan 3.0 recently topped the Artificial Analysis video editing leaderboard and sits in the top three for text-to-video. FLUX 3 Video has built a reputation for photographic fidelity. Both have active communities posting prompts, failure cases, and workflow notes. Grok Imagine Video 1.5 Agent enters with a competitive score but none of that surrounding knowledge. For a filmmaker, that gap matters: a model you understand imperfectly is often less useful than a slightly lower-scoring model you have learned to drive.

The comparison against Seedance 2.5 is notable in a different way. Seedance has been the go-to for consistent character work and realistic footage aesthetics over the past few weeks, and Grok Imagine Video 1.5 Agent now scores above it. Whether that translates into better character consistency or cinematic output quality is entirely unverified at this point.

What to test first

If and when access opens, the most useful first test is a direct comparison on a scene type you already have a strong Wan 3.0 or Seedance 2.5 prompt for. Run the same prompt through Grok Imagine Video 1.5 Agent and look at motion quality, subject consistency across a single clip, and how it handles camera movement instructions. Those three variables will tell you more about practical utility than the Arena score alone. The "Agent" framing also suggests it may handle multi-step or chained generation differently from a single-shot model, so testing a prompt that requires any form of scene structure or temporal continuity is worth prioritising.

SpaceXAI is now officially a top-five AI video lab by human preference. That is the useful fact to carry forward, with the caveat that nearly everything else about this model remains to be discovered.

---
Sources: Arena leaderboard announcement