Aiide AI secures dual direct supply from Alibaba Cloud Bailian and ByteDance Volcano, adding new variables to the domestic model aggregation battle

A Direct Supply Contract Pushes the Relationship Between Aggregators and Cloud Providers One Step Further

On June 17, AI API aggregation service provider Aiide AI released an announcement simultaneously on its official website and community: it has signed official direct supply cooperation agreements with Alibaba Cloud Bailian and ByteDance Volcengine, with registration channels fully open. On the surface, it’s just another aggregator platform obtaining authorization, but at the 2026 time mark, the implications are quite different.

Over the past two years, a wave of players has entered China’s AI API aggregation track, all following a similar logic—pack mainstream models like GPT, Claude, Gemini, DeepSeek, Qwen, GLM, Kimi into a single key, make them compatible with the OpenAI format, and solve developers’ pain points of “one project calling five providers, five sets of authentication, five sets of billing.” The real depth of this track lies in the “upstream”: whether you have an official key, a reverse-engineered account pool, or a relay from another relay directly determines stability and compliance. This time, Aiide secured both major distribution channels for two domestic large models—Alibaba Cloud Bailian and Volcengine—essentially streamlining the upstream chain.

The Alibaba Cloud Bailian Line: GLM and Kimi Suite at 35% Off

Let’s look at Alibaba first. Aiide obtained not regular reseller credentials but “official direct supply”—according to the announcement, mainstream domestic models like GLM and Kimi run on the Bailian platform, and prices are directly set at 65% of the Alibaba Cloud official price.

This discount is worth unpacking. Alibaba Cloud Bailian’s strategy over the past two years has been “the most open large model platform,” a slogan proposed by Zhou Jingren at the 2023 Yunqi Conference. Since then, the platform has successively integrated almost all top domestic models, including Zhipu ChatGLM, Moonshot Kimi, Baichuan, and 01.AI. Recently, Qwen3.7-Plus was launched with a postpaid reasoning price at 80% of standard rates. In other words, Alibaba’s official pricing has already been trending downward, and Aiide’s additional 35% cut on top of that effectively drives down domestic model invocation costs to a notably aggressive level.

For developers, this means—take a concrete example: building an Agent application where the routing layer calls GLM-4.6 for long-form writing tasks, calls Kimi K2 for long-context retrieval, and calls Qwen3.7-Max for complex reasoning. The combined bill from these three at official prices versus using a single Aiide key is not just a small difference; it’s close to one-third cheaper. On a project with monthly call volumes in the tens of millions of tokens, this is real money.

The key factor is stability. Aiide emphasizes “resources are uniformly accessed via Alibaba Cloud Bailian,” implying that your invocation requests ultimately land on Alibaba’s official inference clusters, enjoying the same SLA as directly signed Alibaba clients. This contrasts sharply with low-cost APIs built on gray account pools—which might work today but collapse tomorrow due to risk control measures—offering genuine assurances in compliance and stability.

The Volcengine Line: ¥10 Million Minimum Guarantee + Official Connection to Seedance 2.0

On ByteDance’s side, the numbers are even more striking. Aiide disclosed that Volcengine provided a ¥10 million guaranteed resource allocation, supporting unlimited concurrent calls, and the Seedance 2.0 video generation model is also officially connected.

What does a ¥10 million minimum guarantee mean? Volcengine’s annual revenue exceeded ¥10 billion in 2024 and is targeting ¥20 billion in 2025. Signing a contract with such a minimum guarantee suggests Aiide has at least reached mid-tier status in Volcengine’s customer segmentation. Volcengine’s most valuable resources now are Doubao large models and Seedance video generation—the former managed to grab clients like Weibo and vivo from Alibaba Cloud by dropping inference input prices to 1% of the industry average; the latter, Seedance 2.0, was just fully opened via API in June at the Force conference, signaling ByteDance’s flagship push into the AI video generation sector.

Seedance 2.0’s notable distinction is that, unlike many video generation APIs that limit concurrency or duration, Volcengine’s official demo already produces 5-second clips with complex camera movements and high-detail lighting. However, video generation consumes dozens of times more computing power than text models, so maintaining stable output under high concurrency is key to whether an aggregator dares to sell it. Aiide’s explicit claims of “sufficient output capacity to meet high concurrency,” along with unlimited concurrency promises, target small and medium-sized teams engaged in mass AI short video production—teams whose nightmare is waiting three hours during peak rendering times.

Claude Max Account Pools and Official Keys: The Aggregator’s Other Leg

For domestic models, Aiide takes the “official cloud supply” route, while for overseas models, it uses a different approach: Claude Max account pools + Claude official keys, with resources provided by offline contracted partners. Gemini and GPT are also supported.

Here’s a translation for developers: Claude, due to Anthropic’s regional restrictions and account termination policies, has been a headache in China—direct connections don’t work, maintaining official accounts is costly, and Max subscriptions have concurrency limits. Players who can stably supply Claude either have overseas company entities capable of bulk enterprise API provisioning or rely on Max account pools. Aiide’s statement “offline contracted partners” means the account pool comes from contractual partners, not temporary amateur accounts—reducing the risk of sudden loss for teams relying on Claude for code generation or long-document analysis.

Notably, Aiide distinguishes between Claude official keys and Max account pools. This is a straightforward practice—official keys go through Anthropic’s official billing, have the highest stability and compliance, but are expensive; Max account pools are cheaper but involve trade-offs in concurrency and stability. Letting users choose according to their business needs is more transparent than platforms that mix resources of varying quality.

OpenAI Compatibility: The Entry Ticket for Aggregators

Now, from resources to product form: Aiide follows the mainstream route of OpenAI-compatible format—one key, one base_url, change the endpoint and you can call all models. This paradigm is now a standard for Chinese aggregators; OpenAI Hub operates similarly.

For developers, OpenAI compatibility means zero code migration cost:

from openai import OpenAI

client = OpenAI(
    api_key="your-aggregator-key",
    base_url="https://aggregator.example.com/v1"
)

resp = client.chat.completions.create(
    model="glm-4.6",  # or kimi-k2, claude-opus-4, gpt-5, etc.
    messages=[{"role": "user", "content": "..."}]
)

Change the model name and you’re calling a different model. Combined with LangChain, LlamaIndex, and various Agent frameworks, it’s practically plug-and-play. The abstraction isn’t complicated, but its value lies in shifting developers’ focus from “authentication and adaptation” to “business logic”—one reason why aggregation platforms like OpenAI Hub have grown faster than single-model providers over the past year.

The Real Divide in the Aggregator Track: Can You Sign Upstream Deals?

Back to the cooperation itself. Aggregated APIs look low-barrier—set up a gateway, build billing, connect a few upstreams, and you’re in business. But the winners in the long run differ entirely based on upstream:

• Are resources officially supplied? Determines compliance and stability
• Can you get cooperative discounts? Determines gross margins and retail competitiveness
• Minimum guarantee and concurrency limits? Determines ability to serve large clients
• New model rollout pace? Determines appeal to developers

By securing both Alibaba Cloud Bailian and Volcengine mainlines, Aiide has connected the two largest domestic model distribution channels in one go. On Alibaba’s side: GLM, Kimi, Qwen suite, plus third-party integrated models; on ByteDance’s side: Doubao series, Seedance video generation, and upcoming new models in ByteDance’s ecosystem. Such “dual cloud direct supply” resource structures are still relatively rare among domestic aggregators.

Incidentally, OpenAI Hub (openai-hub.com) has a similar approach—package GPT, Claude, Gemini, and DeepSeek into the OpenAI format; one key for all, domestic direct connection, near-zero integration cost. On the domestic model side, OpenAI Hub has already bridged mainstream APIs. Future competition in the aggregation track will likely hinge on extremes in “resource stability” and “price competitiveness.”

Dragon Boat Festival 20% Off Recharge: A Signal Beyond Promotion

One more note—Aiide is offering a Dragon Boat Festival limited-time 20% off recharge from June 17 to June 21. The promotion itself isn’t big news, but when combined with “65% of Alibaba’s official price + 20% off recharge,” it’s effectively 52% of Alibaba’s official price. Such hefty discounts are short-term acquisition tactics; whether they can be sustained depends on when Volcengine’s ¥10 million guarantee runs out and whether Alibaba’s cooperation discounts renew.

Advice for developers: run small-scale tests focusing on three key metrics—

1. Latency and concurrency: especially for heavy models like Seedance 2.0—check peak period queue times
2. Billing transparency: whether token counting matches official records and if there’s hidden markup
3. Failure recovery: whether the aggregation layer auto-failovers when upstream jitters

If these three pass, the aggregator is genuinely ready for production use.

In Closing

Changes in domestic model distribution chains over the past two years boil down to a redefinition of the relationship between cloud providers and aggregators. Alibaba Cloud takes the “most open” route, pulling all third-party models into Bailian; Volcengine follows the “self-developed + extreme cost-performance” route, leveraging Doubao series and Seedance to win clients. Different strategies, but both need distribution channels—and aggregators like Aiide sit neatly between cloud providers and end developers, reaping the middle-layer benefits.

How this game plays out next depends on two key factors: whether Anthropic and OpenAI further tighten policies (Claude account pool stability remains a sword hanging over all aggregators) and whether domestic models can truly match GPT-5 and Claude Opus in capabilities (if they can, the value of domestic resources in aggregators’ hands will rise significantly). Either outcome favors developers—more choices, lower prices, minimal migration costs—and this is the true meaning of the aggregation track’s existence.

References

• linux.do - Aiide AI x Alibaba Cloud Bailian x ByteDance Volcengine Joint Announcement: Original Aiide AI announcement with cooperation details and Dragon Boat Festival promotion information