Unpopular Opinion: I’m glad Sora is gone
A viral critique argues Sora's resource drain hindered text model development like GPT-5.4.
A provocative online opinion is gaining traction, arguing that OpenAI's decision to sunset its Sora video generation model is a net positive for the AI ecosystem. The author, identifying as a creative professional, claims Sora failed to deliver reliable results for both hobbyist and commercial work, often being outperformed by competitors like Higgsfield and Google's Veo. The core argument is that Sora represented a massive misallocation of scarce computational resources, which could now be redirected to accelerate development of OpenAI's flagship text models, specifically the anticipated GPT-5.4 and the rumored "Spud" (5.5).
The post outlines a clear wishlist for this newly freed-up capacity. Priorities include boosting compute for next-gen text models, making advanced models like a hypothetical GPT-5.4 Pro available to Plus subscribers with query limits, and enhancing core capabilities like context window length and reasoning accuracy. The author also points to Anthropic's Claude as a benchmark for better cross-app integration, suggesting OpenAI should improve its hooks into professional suites like Microsoft Office and communication platforms. This critique taps into a broader industry debate about whether the race for flashy multimodal demos comes at the expense of refining the foundational, commercially critical text-based reasoning that powers most enterprise AI applications today.
- Author claims Sora was unreliable for professional use, citing Higgsfield and Google's Veo as sometimes superior alternatives.
- Central thesis: Sora's high compute cost diverted resources from core text model development like GPT-5.4 and "Spud" (5.5).
- Wishlist includes: more compute for text models, GPT-5.4 Pro for Plus users, larger context windows, and better app integrations akin to Claude.
Why It Matters
Highlights the strategic tension in AI labs between pioneering new modalities and refining the core text models that drive most real-world business value.