The model, introduced by Alibaba, is positioned as a new bar for coding and cowork; shares of the company jumped.
cottonbro studio
Alibaba has unveiled an AI model called Qwen 3.8-MAX, according to a report. The model is described as a new bar for coding and cowork.
Following the announcement, Alibaba shares jumped, as reported.
The related Hacker News item has 16 points.
The related Hacker News item has zero comments, according to the report.
The new model features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to manage costs. Capable of processing up to one million tokens and multimodal content including video, Qwen3.8-Max has achieved second place worldwide in visual content analysis rankings on the Arena.AI platform. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10.
The new Qwen3.8-Max model features 2.4 trillion parameters and a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to manage costs. Capable of processing up to one million tokens and integrating text, image, and video, the model reached second place globally in visual content analysis rankings. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10.
Qwen3.8-Max features 2.4 trillion total parameters using a mixture-of-experts architecture that activates approximately 95 billion parameters per operation, allowing it to process up to one million tokens including text, image, and video. The model has achieved the highest ranking among China-based text models on Arena.AI and holds second place globally for visual content analysis, notably outperforming Fable5 and GPT5.6 Sol in certain tests. Alibaba plans to make the model available on its Alibaba Cloud Model Studio starting the week of August 10 and will release the open model weights for Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to manage costs. The multimodal model can process up to one million tokens in a single pass and has ranked as the highest text model based in China on the Arena.AI platform. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10 and will release the open model weights for Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope.
The Qwen3.8-Max model features 2.4 trillion parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to optimize costs. Capable of processing up to one million tokens and integrating text, image, and video content, the model has ranked highest among China-based text models on Arena.AI and secured second place globally for visual content analysis. Alibaba plans to make the model available on its Alibaba Cloud Model Studio starting the week of August 10 and has announced that the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B will be released next week on Hugging Face and ModelScope.
The Qwen3.8-Max model features 2.4 trillion parameters using a mixture-of-experts architecture that activates approximately 95 billion parameters per operation, and it can process up to one million tokens in a single pass. According to Arena.AI rankings, the model holds the highest position among China-based text models and ranks second worldwide for visual content analysis, though it trailed behind some Anthropic systems in general text rankings. Alibaba plans to make the model available on its Alibaba Cloud Model Studio starting the week of August 10 and announced that the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B will be released next week on Hugging Face and Modelisco
Qwen3.8-Max features 2.4 trillion total parameters using a mixture-of-experts architecture that activates 95 billion parameters per operation to optimize costs. The multimodal model can process up to one million tokens in a single pass, enabling it to analyze various data types including text, images, video, and live streams. Additionally, Alibaba announced that the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B version will be released next week on Hugging Face and ModelScope.
Qwen3.8-Max is a multimodal model with 2.4 trillion total parameters that utilizes a mixture-of-experts architecture to activate approximately 95 billion parameters per operation. The model features a one-million-token capacity and can process text, image, and video content, even recreating software applications from screenshots. Alibaba plans to release the open weights for both Qwen3.8-Max and its smaller version, Qwen3.8-27B, next week on Hugging Face and ModelScope, with availability on the Alibaba Cloud Model Studio platform starting the week of August 10.
Qwen3.8-Max features 2.4 trillion total parameters using a mixture-of-experts architecture that activates 95 billion parameters per operation, and it can process up to 1 million tokens or approximately 750,000 words per query. The multimodal model, which ranks second globally in visual content analysis, is capable of processing text, images, and video, and is scheduled to be available on the Alibaba Cloud Model Studio platform during the week of August 10. Furthermore, Alibaba plans to release the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B version next week on Hugging Face and ModelScope.
The Qwen3.8-Max model features 2.4 trillion total parameters using a mixture-of-experts architecture that activates 95 billion parameters per operation, and it can process up to 1 million tokens, equivalent to approximately 750,000 words. Multimodal capabilities allow the model to analyze text, images, and video, with specific applications including recreating websites from screenshots, generating interactive games, and converting 2D floor plans into 3D visualizations. Alibaba plans to release the open weights for both Qwen3.8-Max and its smaller Qwen3.8-27B version next week, with the former scheduled to become available on the Alibaba Cloud Model Studio platform during the week of August 1
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to manage processing costs. The multimodal model can process up to one million tokens, enabling the analysis of up to 750,000 words, images, and video content in a single pass. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10 and will release the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B version next week on Hugging Face and ModelScope.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates 95 billion parameters per operation to manage costs. The multimodal model, which can process up to one million tokens or approximately 750,000 words, is capable of analyzing text, images, and video content. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10 and will release the open weights for both Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope.
The model features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates 95 billion parameters per operation to manage costs. Capable of processing up to one million tokens, including text, images, and video, Qwen3.8-Max has ranked second globally on the Vision Arena and fifth on the Text Arena. Alibaba plans to release the open weights for both this model and the smaller Qwen3.8-27B next week, with availability on the Alibaba Cloud Model Studio platform expected the week of August 10.
Qwen3.8-Max features 2.4 trillion total parameters using a mixture-of-experts architecture that activates 95 billion parameters per operation to manage costs. The multimodal model can process up to 1 million tokens, including text, images, video, and even 100-hour livestreams, enabling tasks such as 3D modeling and reconstructing websites from screenshots. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform starting the week of August 10 and will release the open weights for both Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation. The multimodal model can process up to one million tokens, enabling it to analyze extensive documents, 100-hour livestreams, and even reconstruct websites from screenshots. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform starting the week of August 10 and will release its open weights next week on Hugging Face and ModelScope.
Qwen3.8-Max features a 2.4 trillion parameter architecture using a mixture-of-experts method that activates 95 billion parameters per operation to manage costs. The multimodal model can process up to 1 million tokens, enabling it to analyze long documents, 100-hour livestreams, and even recreate websites from screenshots. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform starting the week of August 10 and will release its open weights next week.
The model features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates 95 billion parameters per operation. Capable of processing up to 1 million tokens, including text, images, and video, it can handle tasks such as analyzing 100-hour livestreams or reconstructing websites from screenshots. Additionally, Alibaba plans to release the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B next week, with the Max model becoming available on the Alibaba Cloud Model Studio platform during the week of August 10.
The 2.4 trillion parameter model features a mixture-of-experts architecture that activates 95 billion parameters per operation and can process up to 1 million tokens, equivalent to 750,000 words. Alibaba plans to make Qwen3.8-Max available on its Alibaba Cloud Model Studio platform starting the week of August 10, while also announcing that the open weights for both Qwen3.8-Max and the smaller Qwen3.8-27B will be released next week on Hugging Face and ModelScope. Additionally, the company's shares rose 4.5% in New York premarket trading and 7% on the Hong Kong exchange following the announcement.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation to manage processing costs. The multimodal model can process up to 1 million tokens, enabling the analysis of content such as 100-hour livestreams or extensive software codebases. Alibaba plans to make the model available on the Alibaba Cloud Model Studio platform during the week of August 10 and will release the open weights for both Qwen3.8-Max and Qwen3.8-27B next week.
Alibaba disclosed that Qwen 3.8-Max features a 2.4 trillion parameter mixture-of-experts architecture, activating 95 billion parameters per operation to balance performance and cost. The multimodal model supports a one-million-token context window, enabling the processing of lengthy documents, 100-hour livestreams, and complex software projects, which the company claims the model autonomously managed during a 16-day trial. Additionally, Alibaba announced plans to release the open model weights for both Qwen 3.8-Max and a smaller 27B version next week, while the platform QwenWork has entered public beta.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation. The model is capable of processing up to 1 million tokens, or roughly 750,000 words, in a single pass and is designed to handle multimodal inputs including text, images, and video content. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week starting August 10, while also announcing that it will release the open weights for both Qwen3.8-Max and its smaller Qwen3.8-27B version next week.
Qwen3.8-Max features 2.4 trillion total parameters using a mixture-of-experts architecture that activates 95 billion parameters per operation, while its capacity allows for processing up to 1 million tokens or 750,000 words in a single pass. The model has achieved significant rankings, placing second on the Vision Arena and fifth on the Text Arena leaderboard. Additionally, Alibaba plans to release the model's open weights next week and will make it available on the Alibaba Cloud Model Studio platform starting the week of August 10.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates approximately 95 billion parameters per operation. The model is a multimodal system capable of processing up to 1 million tokens, enabling it to analyze 100-hour livestreams, reconstruct websites from screenshots, and create interactive games from natural-language instructions. Alibaba plans to make the model available on its Alibaba Cloud Model Studio platform during the week of August 10 and will release the open weights for both Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope.
Alibaba has announced that the open weights for Qwen3.8-Max and the smaller Qwen3.8-27B version will be released next week on Hugging Face and ModelScope. The new model is priced at $2 per million input tokens and $6 per million output tokens, which is nearly 30% of what Claude Fable 5 charges. Additionally, Alibaba plans to make Qwen3.8-Max available on its Alibaba Cloud Model Studio platform during the week of August 10.
Qwen3.8-Max features 2.4 trillion total parameters and utilizes a mixture-of-experts architecture that activates 95 billion parameters per operation to optimize costs. The model offers a context capacity of up to 1 million tokens, enabling it to process extensive data like 100-hour livestreams or large software codebases. Additionally, Alibaba plans to release the open weights for both Qwen3.8-Max and its smaller Qwen3.8-27B version next week on Hugging Face and ModelScope.
Alibaba revealed that Qwen3.8-Max, with 2.4 trillion total parameters, activates only 95 billion per request via a sparse mixture-of-experts architecture to reduce costs, and is priced at $2 per million input tokens and $6 per million output tokens — nearly 30% cheaper than Claude Fable 5 — while also becoming the first Max-scale Qwen model released with open weights, available on Hugging Face and ModelScope next week.
Alibaba revealed that Qwen3.8-Max, with 2.4 trillion total parameters, activates only 95 billion per operation via a sparse mixture-of-experts architecture, enabling it to process up to 1 million tokens—equivalent to 750,000 words—per query while reducing costs to nearly 30% of Anthropic’s Fable 5, at $2 per million input and $6 per million output tokens; the model also autonomously completed a 16-day software project, producing 265 commits and 151 issues without human intervention, and will release its open weights next week on Hugging Face and ModelScope.
Alibaba revealed that Qwen3.8-Max, with 2.4 trillion total parameters but only 95 billion activated per request via a sparse mixture-of-experts architecture, can process up to 1 million tokens per query—enabling analysis of entire software codebases or 100-hour livestreams in one go—and is now accessible via QwenCloud and Qwen Studio, with open weights set for release next week on Hugging Face and ModelScope; internal tests show it autonomously completed a 16-day software project, outperformed Fable 5 and GPT-5.6 Sol in key benchmarks, and operates at nearly 30% the cost of Claude Fable 5, while ranking second globally in multimodal tasks and highest among Chinese models on Arena.AI's text,
Alibaba revealed that Qwen3.8-Max, with 2.4 trillion total parameters, activates only 95 billion per request via a sparse mixture-of-experts architecture, enabling it to process up to 1 million tokens per query—equivalent to 750,000 words—and perform autonomous software projects over 16 days, including 265 commits and 151 issues without human input, while its pricing at $2/$6 per million input/output tokens is nearly 30% cheaper than Anthropic’s Fable 5 and undercutting Moonshot AI’s Kimi K3, which uses 104.2 billion activated parameters.
Qwen3.8-Max, which was previously described only as a new coding-focused AI model, is now confirmed to have 2.4 trillion total parameters—approaching Moonshot AI’s Kimi K3 (2.8 trillion)—and activates only 95 billion per request via a sparse mixture-of-experts architecture, reducing costs to nearly 30% of Anthropic’s Fable 5; it also becomes the first Max-class Qwen model released with open weights, available on Hugging Face and ModelScope next week, while its pricing at $2/M input and $6/M output tokens and 1M-token context length position it as a cost-efficient, multimodal agent for long-term autonomous tasks like full software project development, legal document analysis, and video-to-kb
