Prem Akkaraju, former CEO of Weta Digital, leads Stability AI after Emad Mostaque resigned to pursue decentralized AI. Mostaque’s history of making grandiose statements contributed to the departure of key engineers and senior executives, which followed reports of the company facing $30 million in losses against $5 million in revenue during early 2024. Executive Chairman Sean Parker joined the leadership team alongside Akkaraju. This team manages an $80 million investment from investors including Coatue, Lightspeed Venture Partners, and Eric Schmidt. The company focuses on enterprise licensing and API services to generate revenue. In September 2026, hiring at the company accelerated with three new job postings. Stability AI employs approximately 189 people as of October 19, 2026. Mostaque previously led a team that secured a $101 million seed round and reached a $1 billion valuation. The company formed a strategic partnership with WPP in March 2025 to expand its reach. The departure of researchers like Robin Rombach, Andreas Blattmann, and Dominik Lorenz previously slowed development, but the company has since secured $125 million in total investment to stabilize its research efforts.
The video generation hierarchy
Stable Video 3 enters the generative video market to compete with closed-API leaders like Sora 2 and Runway Gen-4. Sora 2 wins the cinematic category, capturing 42 of 50 reviewer comparisons because it provides superior lighting, volumetric atmosphere, and realistic reflections. Runway Gen-4 wins the product/marketing category with 30 of 50 wins because of its motion brushes and keyframing tools. Stable Video 3 dominates animation-stylized prompts with 33 of 50 wins because creators frequently apply community LoRAs to its open weights. The model also wins social-vertical prompts with 28 of 50 wins. Sora 2 holds a lead in hard-physics tasks, winning 38 of 50 prompts through a physics-aware backbone. A patch released on April 28, 2026, improved hand consistency and reflection rendering in Stable Video 3.1. The model produces 12-second clips at 1080p, though chained workflows allow 24-second extensions. Sora 2 supports 20 seconds at 4K, while Runway Gen-4 supports up to 16 seconds at 1080p with high temporal consistency. Sora 2 struggles with social-vertical prompts and captured only 16 of 50 wins because it produces excessive cinematic gravitas instead of casual phone footage. Can Stable Video 3 overcome the physics advantage held by Sora 2?
Production costs and hardware
Creators choose between hosted APIs and self-hosted hardware based on volume. Stable Video 3 functions at an amortized cost of $0.05 to $0.10 per second on owned hardware like an RTX 5090. Running the model on rented infrastructure like Lambda Labs costs roughly $0.10 to $0.20 per second. In contrast, Sora 2 costs $0.60 to $0.80 per second at standard settings via OpenAI API. Runway Gen-4 costs $0.24 to $0.36 per second following its April 30 price cut. Users often find Runway’s credit consumption unpredictable when they iterate through complex scenes. Full-quality Stable Video 3 requires 32GB of VRAM, though a 16GB quantized profile produces 80% of the quality for those with less power. Users with Apple Silicon require 64GB of unified memory for the full profile.
| Model | Max Video Length | Primary Strength | Hardware Requirement |
|---|---|---|---|
| Sora 2 | 60 seconds | Cinematic realism | API-only |
| Runway Gen-4 | 16 seconds | Directorial control | API-only |
| Stable Video 3 | 12 seconds | Stylization and cost | 32GB VRAM |
Stability AI allows local use and fine-tuning under a permissive commercial license. This license permits the redistribution of fine-tunes and the integration of models into commercial products. The license avoids the watermark requirements tied to Sora 2’s free-tier. Professionals often use Stable Video 3 for high-volume work and Sora 2 for hero pieces. Runway’s plans include the Standard tier at $15, the Pro tier at $35, and the Max tier at $95. You should evaluate your monthly credit burn before committing.
