Marina Lyubimova
⬤ Artificial Analysis published a new comparison of leading AI video generation models, and P-Video from Pruna AI came out on top as both the fastest and lowest-cost option among models tested through public APIs. The model generates a 720p five-second video in roughly 10 seconds, putting it in the most favorable position on a chart that maps API generation time against API price — a spot that signals real practical value for developers building at scale.
⬤ On pricing, P-Video comes in at $1.20 per minute for 720p and $2.40 per minute for 1080p output. That makes it noticeably cheaper than rivals like LTX-2 Fast from Lightricks at $2.40/min and MiniMax Hailuo 2.3 at $2.80/min. Per-second, that's around $0.02 for 720p and $0.04 for 1080p — figures that put it among the most affordable AI video systems on the market right now.
⬤ The comparison also covered several other major video models, including OpenAI's Sora 2, ByteDance's Seedance 1.5 Pro, Wan 2.5 Preview, and Kling 3.0 models available through fal.ai. Most of these offer higher-resolution output or more polished quality, but come with longer generation times or steeper API costs — the classic trade-off developers have to weigh when picking video infrastructure.
⬤ P-Video is available through Pruna AI's API and integrated across platforms including Replicate, Runware, Cloudflare, Runpod, Wiro AI, and Qubrid AI. It currently ranks #41 in Text-to-Video and #42 in Image-to-Video on the Artificial Analysis Video Arena — placing it in a performance tier alongside Seedance 1.0 Mini and Wan 2.2 A14B. The broader takeaway: as cost and latency keep dropping, real-time AI video creation is moving from niche experiment to production-ready infrastructure.
Marina Lyubimova
