OctaFuse this week updated its self-hosted AI gateway to 2.8.0, and one change is worth flagging for businesses wiring AI into their operations: Qwen Image 3.0, Qwen Image 3.0 Pro, Wanxiang 2.7 Image, and Wanxiang 2.7 Image Pro — four domestic image models — can now be invoked directly through OpenAI's general image generation API, no longer requiring separate integration with Alibaba Bailian's proprietary protocol.
What this is
OctaFuse is a category of tool known as an "AI API gateway" — picture an "AI traffic master switch" installed inside the enterprise. When a company runs large language models, image models, and voice models side by side, this master switch handles unified access, quota management, routing, and billing. OctaFuse is one such option, open source and designed to be self-hosted in your own data center or on a cloud VM.
2.8.0 does three main things: subscriptions and top-ups are now split into two separate ledgers that can run concurrently — once the monthly plan runs out, top-ups are drawn down automatically; routing configuration shifts from hand-typed protocols to dropdown menus, removing a class of misconfiguration bugs; and the four domestic image models above now support the OpenAI protocol. The first two make enterprise books cleaner; the third means code written against OpenAI's API can switch to domestic models with almost no changes.
Industry view
Supporters read this as a small piece of evidence for the thesis that "OpenAI's protocol has become a de facto industry standard." Anthropic, Google, and Mistral are all nudging their models toward OpenAI compatibility; Alibaba Bailian this time backs the standard indirectly at the gateway layer, effectively ratifying it. Companies that already have OpenAI-style calls wired into overseas workflows will face materially less engineering when swapping in domestic models.
There is pushback too. A domestic AI integrator told us that self-hosted gateways are essentially "protocol translators" — their value erodes as major vendors ship their own compatible endpoints, and Alibaba Bailian has, in fact, already published its own OpenAI-compatible endpoint publicly. Self-hosting also means the enterprise owns upgrades, security, and capacity; for SMBs without dedicated ops staff the math doesn't pencil out, and it reads more as a transitional choice than a destination.
Impact on regular people
For enterprise IT: multi-model mixing and unified billing are becoming genuinely viable, but self-hosting still demands at least one engineer who knows containers and APIs. Outsourcing to an integrator is the more common starting posture.
For working professionals: colleagues using AI to generate images or draft copy won't feel any difference in the short term. The real relief lands on engineers writing code and building product prototypes, who hit far fewer "protocol doesn't talk to protocol" walls.
For the consumer market: gateways are infrastructure — they don't show up in consumer products directly. If they truly compress enterprise AI costs, the indirect effect will surface gradually over the next year or two in the pricing of AI customer service and design SaaS.