Key specifications:

• 1 million token context window, enabling processing of documents up to 750,000 words in a single pass
• Ranked #5 in Text Arena and #2 in Vision Arena on independent benchmarks
• Made available globally via API on Alibaba Cloud Model Studio, with model weights promised for open-source release around August 10
• Alibaba shared results showing it delivered comparable or sometimes better scores than Anthropic's Fable 5
• According to the company, the model successfully completed a 16-day coding project without human input
• Pricing was set significantly lower than comparable frontier models, with reports suggesting a strategy of aggressive cost competition against US rivals

This is the first Max-class model in the Qwen series to go open-weight, putting a frontier-scale system permanently beyond the reach of export controls once downloaded .

QwenWork — August 3, 2026

Simultaneously, Alibaba launched QwenWork, an enterprise workplace AI agent platform that consolidates three previously separate agent products — QoderWork, MuleRun, and Wukong — under a single platform overseen by DingTalk's CEO . Key details:

• Launched in public beta on the same day as Qwen3.8-Max
• Positioned as an end-to-end enterprise agent solution integrated with Alibaba's existing business software ecosystem, including DingTalk which serves more than 20 million enterprises and organizations
• Available through a cloud-based web interface and a desktop client with direct access to local computers
• Early reports noted it brings China-specific regulatory considerations (state-law risk) into enterprise AI workflows

Wan3.0 Video Model — August 6, 2026 (Public Beta)

Alibaba's Tongyi Lab released Wan3.0 in public beta on August 6, 2026, via Qwen Cloud and the DashScope API . Major features include:

• Native 30-second single-pass video generation — up from 2-15 seconds in Wan 2.7, eliminating the need to stitch clips together
• Omni-Reference input system — accepts documents (PDF, DOC, XLS, PPT, MD, TXT, KEY, Pages), spreadsheets, slide decks, and live URLs alongside traditional text, image, audio, and video inputs; an industry first for document-to-video generation
• Output at up to 1080p resolution (no 4K tier, despite pre-launch rumors)
• Built on a Diffusion Transformer (DiT) architecture
• Pricing: $0.05 per second at 480p, $0.10 at 720p, $0.20 at 1080p. A full 30-second generation therefore costs $1.50, $3, or $6
• Unifies text-to-video, image-to-video, reference-based generation, video editing, motion driving, and video replication into a single model, versus separate models in Wan 2.7
• Supports up to 20 references across image, video, audio, documents, and webpages, held consistent across the clip

Qwen Open Platform (Qianwen Open Platform) — August 10, 2026

On August 10, Alibaba launched the Qianwen Open Platform, allowing third-party developers to build custom AI agents integrated across smartphones, PCs, and AI glasses . The first group of partners spans more than 10 sectors including logistics, real estate, and wealth management, with participants including SF Express, home-rental platform Ziroom, and Midea Group . This opened the Qwen ecosystem beyond Alibaba's own products, positioning Qwen as an all-in-one application designed to carry out complex real-world tasks .

Qwen AI Arena — August 12-13, 2026

Alibaba Cloud launched Qwen AI Arena (千问AI Arena), a challenge and evaluation platform for AI agents that tests them against real-world business scenarios . Key aspects:

• Provides developers with models, runtime environments, and evaluation tools to submit and test agent solutions
• Results are displayed on public leaderboards, with top-performing solutions eligible for inclusion in the Qwen AI ecosystem
• Inaugural challenge — co-launched with Aidge — focuses on cross-border e-commerce product listing scenarios
• Qwen AI Arena introduces an innovative Agent automated evaluation capability combined with expert review: automated testing began in mid-August, with the top 30 submissions moving to expert review
• Alibaba said new challenges across different industries and scenarios will be added over time

Summary: A Coordinated 10-Day Push

Alibaba's early-August 2026 releases formed a coordinated push across the full AI stack: a frontier foundation model (Qwen3.8-Max), an enterprise agent platform (QwenWork), a developer ecosystem play (Qwen Open Platform), a video generation tool (Wan3.0), and a community-driven agent benchmarking system (Qwen AI Arena) — all within roughly 10 days. The strategy combines increasingly capable models with lower-cost access and distribution through Alibaba's existing business software .

Key specifications:

• 1 million token context window, enabling processing of documents up to 750,000 words in a single pass
• Ranked #5 in Text Arena and #2 in Vision Arena on independent benchmarks
• Made available globally via API on Alibaba Cloud Model Studio, with model weights promised for open-source release around August 10
• Alibaba shared results showing it delivered comparable or sometimes better scores than Anthropic's Fable 5
• According to the company, the model successfully completed a 16-day coding project without human input
• Pricing was set significantly lower than comparable frontier models, with reports suggesting a strategy of aggressive cost competition against US rivals

This is the first Max-class model in the Qwen series to go open-weight, putting a frontier-scale system permanently beyond the reach of export controls once downloaded .

QwenWork — August 3, 2026

Simultaneously, Alibaba launched QwenWork, an enterprise workplace AI agent platform that consolidates three previously separate agent products — QoderWork, MuleRun, and Wukong — under a single platform overseen by DingTalk's CEO . Key details:

• Launched in public beta on the same day as Qwen3.8-Max
• Positioned as an end-to-end enterprise agent solution integrated with Alibaba's existing business software ecosystem, including DingTalk which serves more than 20 million enterprises and organizations
• Available through a cloud-based web interface and a desktop client with direct access to local computers
• Early reports noted it brings China-specific regulatory considerations (state-law risk) into enterprise AI workflows

Wan3.0 Video Model — August 6, 2026 (Public Beta)

Alibaba's Tongyi Lab released Wan3.0 in public beta on August 6, 2026, via Qwen Cloud and the DashScope API . Major features include:

• Native 30-second single-pass video generation — up from 2-15 seconds in Wan 2.7, eliminating the need to stitch clips together
• Omni-Reference input system — accepts documents (PDF, DOC, XLS, PPT, MD, TXT, KEY, Pages), spreadsheets, slide decks, and live URLs alongside traditional text, image, audio, and video inputs; an industry first for document-to-video generation
• Output at up to 1080p resolution (no 4K tier, despite pre-launch rumors)
• Built on a Diffusion Transformer (DiT) architecture
• Pricing: $0.05 per second at 480p, $0.10 at 720p, $0.20 at 1080p. A full 30-second generation therefore costs $1.50, $3, or $6
• Unifies text-to-video, image-to-video, reference-based generation, video editing, motion driving, and video replication into a single model, versus separate models in Wan 2.7
• Supports up to 20 references across image, video, audio, documents, and webpages, held consistent across the clip

Qwen Open Platform (Qianwen Open Platform) — August 10, 2026

On August 10, Alibaba launched the Qianwen Open Platform, allowing third-party developers to build custom AI agents integrated across smartphones, PCs, and AI glasses . The first group of partners spans more than 10 sectors including logistics, real estate, and wealth management, with participants including SF Express, home-rental platform Ziroom, and Midea Group . This opened the Qwen ecosystem beyond Alibaba's own products, positioning Qwen as an all-in-one application designed to carry out complex real-world tasks .

Qwen AI Arena — August 12-13, 2026

Alibaba Cloud launched Qwen AI Arena (千问AI Arena), a challenge and evaluation platform for AI agents that tests them against real-world business scenarios . Key aspects:

• Provides developers with models, runtime environments, and evaluation tools to submit and test agent solutions
• Results are displayed on public leaderboards, with top-performing solutions eligible for inclusion in the Qwen AI ecosystem
• Inaugural challenge — co-launched with Aidge — focuses on cross-border e-commerce product listing scenarios
• Qwen AI Arena introduces an innovative Agent automated evaluation capability combined with expert review: automated testing began in mid-August, with the top 30 submissions moving to expert review
• Alibaba said new challenges across different industries and scenarios will be added over time

Summary: A Coordinated 10-Day Push

Alibaba's early-August 2026 releases formed a coordinated push across the full AI stack: a frontier foundation model (Qwen3.8-Max), an enterprise agent platform (QwenWork), a developer ecosystem play (Qwen Open Platform), a video generation tool (Wan3.0), and a community-driven agent benchmarking system (Qwen AI Arena) — all within roughly 10 days. The strategy combines increasingly capable models with lower-cost access and distribution through Alibaba's existing business software .

• Developers must build an Agent that generates complete listing materials (product copy, images, detailed images) for US, South Korea, and Brazilian markets in a single run, including output in English, Korean, and Portuguese
• The Qwen AI platform provides participants with the required token quotas and open access to multimodal model capabilities for text, image, and video generation

• Developers must build an Agent that generates complete listing materials (product copy, images, detailed images) for US, South Korea, and Brazilian markets in a single run, including output in English, Korean, and Portuguese
• The Qwen AI platform provides participants with the required token quotas and open access to multimodal model capabilities for text, image, and video generation