The Extended Brief

Alibaba Targets Coding and Office Tasks With Low-Cost AI Model

Brief by The AI News AI newsroom · Aug 26, 2026, 11:24 AM EDT edition

Original reporting by PYMNTS — AI — PYMNTS · published Aug 26, 2026, 10:41 AM EDT

Alibaba's new open-weight model promises strong coding and office performance at $0.16 per million input tokens, giving cost-sensitive builders a cheaper option to evaluate.

Key points

• Alibaba says Qwen3.8-Flash trained at one-ninth the cost of Qwen3.7-Plus while outperforming it, especially on coding and office tasks. source ↗
• The production version will cost $0.16 per million input tokens and $0.47 per million output tokens via the QwenCloud API. source ↗
• The open-weight multimodal MoE has 125 billion parameters with 6 billion activated per token and previews the Qwen4 architecture. source ↗
• Alibaba's open-weight models reportedly passed 3 billion downloads in six months, ahead of Meta, Google, and DeepSeek. source ↗
• Apple reportedly worked with Alibaba to develop an AI model for China, with Apple Intelligence launching there within months. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Benchmark Qwen3.8-Flash against your current coding and document-workflow models once the QwenCloud API launches, using the published $0.16/$0.47 per-million-token pricing to project cost savings.
• Download the open-weight preview now to test whether the 6B-active-parameter MoE handles your multimodal workloads before committing to the hosted production version.
• If you ship AI features on devices in China, factor in the reported Apple–Alibaba collaboration, since Apple Intelligence there will reportedly use an Alibaba-trained model.

Context

Mixture-of-experts models activate only a small slice of their parameters per token — 6 billion of 125 billion here — which is how Alibaba claims much lower training and inference costs. Open-weight releases let developers download and adapt models freely, and Alibaba says the Qwen family has open-sourced over 460 models with more than 300,000 derivatives. Apple has previously relied on third-party models to power AI features in iPhones sold in China.

What to watch

• Independent coding benchmarks after the production API ships will test Alibaba's claim of beating Qwen3.7-Plus at one-ninth its training cost.
• An Apple Intelligence launch date for China would confirm the reported Alibaba partnership.

Related briefs

• VMs won't contain cyber-capable agents
• OpenAI Jalapeño: Better Than Nvidia Blackwell
• Nvidia to hike prices by 15%, on top of an even larger increase in July
• Are Open Models Catching Up?

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.