xAI's new video model takes top spot at an 86% discount

For months, the biggest issue in generative video hasn't been visual spectacle—it has been the brutal computing economics. Just two months after OpenAI killed off its consumer Sora app citing unsustainable compute costs, xAI is attempting to rewrite the market's pricing structure entirely.

Reports confirmed that xAI has officially pushed Grok Imagine Video 1.5 into full general availability across its API, grok.com, and mobile apps. Along with the launch, the model vaulted to the top of the independent Image-to-Video Arena leaderboard, gaining 52 Elo points over its predecessor to beat out Sora 2, Google's Veo 3.1, and Kling. But the headline that caught every studio's eye is price: at $0.14 per second for 720p output, xAI is undercutting Sora 2 Pro by roughly 86 percent.

Autoregressive video with built-in audio

Grok 1.5's jump on the leaderboard comes courtesy of its Aurora engine, an autoregressive mixture-of-experts architecture that breaks from industry orthodoxy.

While rivals like Sora, Runway, and Kling rely on diffusion—denoising entire video volumes simultaneously across all frames—Aurora treats video like an LLM treats language. It generates frames strictly in sequence, conditioning each new frame on everything that came before. That sequential memory is why camera pans maintain steady trajectories, characters don't melt between cuts, and lighting stays anchored across the entire clip.

Even better for workflow pipelines, audio is synthesized in the exact same inference pass. Ambient noise, sound effects, and lip-synced speech are generated directly with the visuals, eliminating the need to chain a separate audio model downstream. For interactive jobs, xAI also introduced Video 1.5 Fast, which churns out a 6-second clip in about 25 seconds—a 40 percent speed bump.

The trade-off: locked at 720p

The pricing difference is dramatic. At $4.20 per minute ($0.14 per second at 720p), a studio generating 100 minutes of video monthly would save around $2,580 compared to Sora 2 Pro rates ($30 per minute at 1024p), or nearly $1,980 compared to Google's Veo 3.1 Quality tier ($24 per minute).

Yet Aurora's sequential design creates an immediate technical bottleneck: resolution. Jumping from 720p to 1080p multiplies pixel tokens by 2.25 times per frame. Because an autoregressive model must calculate those tokens step-by-step rather than in parallel batches, high resolutions become computationally punishing. As a result, Grok 1.5 is capped at 720p for now, with a hard 15-second generation limit and noticeable degradation after chaining clips two or three times.

There is also platform baggage. Following severe controversies surrounding non-consensual imagery in late 2025, xAI still faces active regulatory scrutiny across multiple jurisdictions, making rigorous content filtering an ongoing pressure point.

Why it matters

For creators and small development studios, Grok Imagine Video 1.5 changes the math on rapid video production. If your deliverables live on social feeds, mobile screens, or early-stage mood boards, 720p is often good enough—and having synchronized audio included out of the box cuts down pipeline complexity significantly. It makes iterative prototyping financially viable for teams that were previously priced out.

For the broader AI industry, xAI is aggressively occupying the vacuum left by OpenAI's consumer retreat and the impending sunset of the Sora 2 API in September. By delivering the top crowd-sourced quality score at a fraction of Google's pricing, xAI is forcing competitors to defend their profit margins.

We think xAI made a remarkably calculated compromise here. Instead of chasing uncompressed 1080p production assets, they doubled down on camera stability, single-pass audio, and raw affordability. It may not replace high-end VFX suites yet, but it looks like the exact practical compromise AI video needed to move from tech demos into routine creative workflows.