China Mobile Cloud and Zhipu signed a strategic cooperation agreement covering domestic compute adaptation, model tuning, and a joint Token development center to improve token production efficiency. Zhipu's GLM models will be integrated into China Mobile Cloud's MoMA model service platform, with plans for joint AI Coding products and finance/government solutions. No investment figures, timelines, or verifiable technical metrics were disclosed.
MaaS
模型即服务
A service model providing pre-trained model inference capabilities via cloud APIs.
Curated Intelligence
IDC data shows China's MaaS market reached 10.6 billion yuan in H1 2026, up nearly 9x year-on-year, with token consumption at 2,857.9 trillion, up over 430%; Google separately reported processing 3,200 trillion tokens monthly. The piece argues that constrained on-robot compute and thermal limits make local embodied-model deployment impractical, pushing cloud-based pay-per-call 'embodied AI MaaS' as the next battleground.
Tencent Cloud announced a major upgrade to its MaaS product line at the Global Digital Ecosystem Conference, shifting focus from simple API calls to controllable task delivery via components like TokenHub, TI-ONE, and Agent Runtime. This move addresses enterprise challenges in multi-model scheduling, governance, and cost accounting, aiming to push AI applications into production environments.
Huawei Cloud will cease sales of openPangu-2.0-Flash/Pro models on Nov 6 and fully decommission them by Nov 30. Users are advised to migrate to third-party alternatives like DeepSeek-V4.1 or GLM-5.3, while new Pangu series models remain in planning without specific details.
Mistral has released the weights for its 1-trillion-parameter model, Mistral Large 4, while cutting API prices by approximately 50%. The model, refined through additional reinforcement learning, outperforms GLM-5.3 on Terminal-Bench but ties it on the Coding Agent Index, with a key advantage in reduced token usage per task. Notably, the licensing terms have shifted from Apache 2.0 to a modified MIT license, imposing stricter commercial requirements on Model-as-a-Service providers.
IDC reports that China's MaaS market grew nearly 9x YoY in H1 2026. SenseTime holds the second position in the private LLM market with a 9.4% share, leveraging Forward Deployed Engineers (FDE) to shift focus from model capabilities to end-to-end platform delivery and measurable business outcomes.
The CAICT released its 2026 MaaS Performance Monitoring Report, revealing that daily token consumption in China reached 500 trillion, nearly quadrupling since late 2025. While average tokens per second (TPS) doubled, time to first token (TTFT) increased due to larger model scales, with significant performance disparities observed between third-party and official platforms.
Monitoring data reveals that China's LLM service prices rebounded significantly in 2026 after a price war, while throughput (TPS) doubled compared to 2025. Despite performance gains, Time To First Token (TTFT) increased due to larger model scales, and significant performance disparities exist between third-party and official platforms.
Alibaba announced a comprehensive upgrade to the Qwen AI platform at the Yunqi Conference, shifting focus to deploying Agents into production. New features include Agent Studio, API Speed Mode, and industry solutions aimed at converting intelligence into productivity. These capabilities are now live for developer integration.
Alibaba announced the upcoming 2026 Yunqi Conference technical forum on September 22 in Hangzhou, focusing on MaaS and Agent strategies. The event plans to unveil updates to Qwen AI Platform, Qoder, and mobile intelligence solutions, but currently offers only a preview with no specific technical parameters or substantive results disclosed.