Alibaba's Qwen team previewed Qwen3.8-Max-Preview on July 19, making it the highest-end preview model publicly available in the Qwen family so far. The model first appeared in Qwen Cloud's Token Plan and in agent products including Qoder and QoderWork, giving developers early access through those products.
Which products launched Qwen3.8-Max-Preview first?
Qwen says Qwen3.8 is launching with an open-weight release planned, while Qwen3.8-Max-Preview is being offered first as an early build. Qwen Cloud's Token Plan page lists the model among its available models, and Qoder's launch notice says it entered the desktop app, JetBrains plugin, CLI, Cloud Agents, and mobile and web apps on July 19.
Qoder is also offering a limited-time promotion for the model: calls consume 10% of the standard Credits rate, while some nighttime periods offer discounts of up to 98%. Eligibility and supported products are governed by Qoder's official campaign terms.
2.4T parameters and complex agentic tasks
Qwen has disclosed a total parameter count of 2.4 trillion for Qwen3.8 and says the model is still evolving. Qoder's documentation says that Qwen3.8-Max-Preview improves on the previous flagship Qwen3.7-Max in coding and professional productivity, covering complex long-horizon tasks such as full-stack development, data analysis, and Office workflows.
Qwen researcher Shuai Bai described it in public comments as the Qwen team's first multimodal model above one trillion parameters. Current disclosures point to image, video, and document understanding, while the architecture, active parameter count, context length, and inference cost remain unpublished. A 2.4-trillion total parameter count alone cannot establish model quality, per-request compute cost, or local deployment requirements.
“Open weights soon,” while the date and license remain unconfirmed
Qwen describes Qwen3.8 as a model that will soon receive an open-weight release, but has not provided a specific date, code repository, license terms, or list of downloadable versions. The currently hosted Qwen3.8-Max-Preview is also not a completed production release; whether the later weights will exactly match this preview build remains open.
Alibaba has not yet published a complete benchmark suite, model card, or reproducible third-party evaluation alongside the announcement. The company's claim that the model is competitive with frontier systems therefore remains a vendor claim at this stage. Developers can form an early view through the listed products, while the model's ranking, deployment requirements, and suitable use cases will need to be reassessed after the weights, technical details, and reproducible evaluations arrive.