
China Mobile's model service platform...
When a system needs text, speech, and multimodal ability together, each vendor does not need a separate integration. MoMA connects once through a unified API gateway, then selects a model per request. Routing can prefer cost, quality, or a balance, and switches on timeout, rate limits, or failure. For data that should not leave the computing environment, confidential models run in confidential containers, with the claim that computation is usable but not visible. Usage is metered by token in real time, with traceable risk control and full-path observation. It fits government and enterprise teams already building on Mobile Cloud who want one place to manage calls to many models.
China Mobile's model service platform MoMA offers a unified API gateway so one integration can call the models on the platform. Xinhua's report on the Mobile Cloud conference says the platform includes China Mobile's Jiutian foundation model plus models such as Qwen, Doubao, and Kimi, covering text generation, speech, and multimodal understanding for government, finance, industry, healthcare, and education. Routing can prefer cost, quality, or a balance, and can switch when a model times out or is rate limited. A later update also matches a model to the scene and can split a complex task across models.
Difficulty: Intermediate
One gateway for the catalog
One integration with the unified API gateway can call the platform's text, speech, and multimodal models, including Jiutian.
Three routing strategies
Routing can prefer cost, quality, or a balance, and moves to an available model on timeout, rate limit, or failure.
Split a complex task
The updated engine can pick a model for the business scene, or split a complex task so different models finish it together.
Confidential containers
Confidential models run in confidential containers, using hardware isolation so data in computation is usable but not visible.
Metering and path observation
Token use is billed as it streams, with traceable risk control and full-path observation of latency, consumption, and resources.
One integration for a government app
A government or finance system connects the gateway once and picks among text, speech, and multimodal understanding per request.
Route for cost or quality
The same kind of task can prefer cost, quality, or a balance, and is redirected when an endpoint fails.
Sensitive data on a confidential model
For data whose computation should not be exposed, use a model in a confidential container and keep an auditable consumption record.
In the report, MoMA uses a unified API gateway so one integration calls models on the platform. Named examples include China Mobile's Jiutian plus Qwen, Doubao, and Kimi, covering text generation, speech, and multimodal understanding. Example scenes are government, finance, industry, healthcare, and education. Which models are open on a given day depends on the Mobile Cloud console.
Routing can prefer cost, quality, or a balance, and switches if a model times out, is rate limited, or fails. A more complex task can be split so different models work on it. For sensitive data, a confidential model in a confidential container is described as usable but not visible during computation. Consumption is metered by token in real time, with a traceable risk-control record.
Real reviews and feedback from users