OpenAI DevDay: agent builder, Sora 2 API at 32¢/sec, and apps inside ChatGPT
Oct 6, 2025
Key Points
- OpenAI released Sora 2 API at 10¢ per second, roughly one order of magnitude cheaper than previous video model pricing, signaling the company prioritizes API distribution over building a standalone consumption platform.
- Agent Builder, a node-based workflow tool, enables prosumers to automate recurring tasks and API calls, raising questions about whether OpenAI can monetize this middle market without alarming subscription consumers.
- ChatGPT hit 800 million weekly active users as OpenAI pursues multiple business models in parallel: consumer apps, API licensing, and hardware ambitions, but the Sora decision suggests the company is betting on embedded tools over owning the end-user experience.
Summary
OpenAI announced four major products at DevDay: Apps inside ChatGPT allow partners like Zillow and Figma to embed interfaces directly into the chat interface, monetizing through their own accounts. Agent Builder is a node-based workflow tool similar to Zapier, capable of scheduling recurring tasks, conditional logic, and chaining API calls for jobs like daily email digests or research reports tied to scheduling. Codex moved to general availability. Sora 2 API costs 10¢ per second, while Sora 2 Pro runs 30¢ per second for longer outputs up to 15 seconds. Those rates are roughly one order of magnitude cheaper than the $3-per-generation pricing OpenAI used for previous video models, though the company likely subsidizes current pricing.
The agent builder announcement immediately prompted posts declaring "OpenAI killed N8N"—the workflow automation platform that has seen organic growth and spawned courses teaching founders how to build businesses on top of it. The concern is legitimate but overstated. The precedent is Google versus Facebook in 2007. Google feared Facebook would disrupt search, ads, calendar, photos (Picasa), YouTube, and other entrenched products. Facebook won in some areas, held some ground in others, and failed to mount credible threats in the rest. The pattern suggests OpenAI will successfully compete in some markets but not all, and incumbents or purpose-built competitors will often retain advantages in distribution, depth, or user behavior.
Releasing Sora 2 via open API signals that OpenAI does not expect to build a consumption platform around video generation. If OpenAI believed Sora had a 30% chance of becoming a major video social network, it would hoard API access to maximize the app's network effects. Instead, it is licensing the model to any video platform—Adobe, CapCut, startups, incumbents—which dilutes Sora's competitive position as a standalone destination. OpenAI is betting Sora succeeds as a creative tool embedded in other platforms, not as a social network or content feed.
This creates tension with ChatGPT's consumer strategy. OpenAI operates two distinct business models: a free consumer product monetized through ads and behind-the-scenes API deals, and an API business selling pure consumption. The agent builder and workflow tools hint at a third, middle market of prosumer consumption. If users set up automated workflows that call GPT-5 Pro or Sora multiple times daily, inference costs could climb to $10, $50, or more per user monthly. That is far higher than a $20 subscription but lower than enterprise spend. OpenAI could move toward consumption-based pricing in this segment, though consumers often prefer all-you-can-eat plans. Whether prosumers—visual artists, newsletter writers, workflow builders—will generate enough recurring API calls to justify OpenAI's infrastructure spend remains unclear, and whether OpenAI can monetize that segment without spooking consumers who fear surprise bills.
ChatGPT hit 800 million weekly active users as of DevDay. Hardware device rumors point to a voice-only pin or assistant with no screen to be released in 2026, but remain constrained by Apple's ecosystem lock-in. iMessage dominance, AirPods integration, and the compounding advantage of Apple's existing wearable installed base make it difficult for any third-party device to compete without messaging interoperability. Humane's failure serves as a warning that technically functional does not equal useful when pitted against an entrenched ecosystem.
OpenAI is conducting multiple campaigns in parallel: infrastructure deals with AMD, Oracle, Broadcom, and TSMC; consumer app launches and API pricing; agent and workflow tooling; and hardware ambitions. The Sora API decision suggests OpenAI is prioritizing API distribution and partner embeds over owning the end-user experience. That is a rational play for a model company competing against Google, but it may foreclose the social network upside Sora could theoretically capture if every generation were automatically surfaced in a feed and ranked. Whether that trade-off proves correct depends on whether prosumer workflows and API licensing generate more value than a TikTok-scale platform could.