Alibaba takes aim at OpenAI and Anthropic with Qwen3.8-Max launch

Alibaba on Monday introduced Qwen3.8-Max, its largest artificial intelligence model to date, expanding its enterprise AI portfolio with an open-weight model designed for software engineering, multimodal reasoning, and other knowledge-intensive business workloads. In a blog post announcing the launch, Alibaba described Qwen3.8-Max as a 2.4-trillion-parameter mixture-of-experts (MoE) model that activates only about 95 billion parameters during inference. The company said the architecture is intended to improve inference efficiency while supporting coding, reasoning and multimodal tasks, with open-weight versions scheduled for release next week through Alibaba Cloud’s Model Studio. “We believe it’s one of the most powerful model available today, compatible to leading frontier AI models, second only to Fable 5,” Alibaba said in an X post. Benchmarks target Anthropic and OpenAI’s coding models Alibaba published internal test results comparing Qwen3.8-Max against Claude Opus 4.8, Claude Fable 5, and OpenAI’s GPT-5.6 Sol on coding benchmarks, including SWE-bench Pro and a proprietary evaluation the company calls NL2Repo-Bench. The company said it evaluated competing models using each vendor’s own coding harness, Claude Code for Anthropic’s models and Codex for GPT-5.6 Sol, and reported the highest published score across available configurations for each rival. Charlie Dai, vice president and principal analyst at Forrester, said the launch signals Alibaba is closing ground on proprietary leaders, though that isn’t the full picture. “Alibaba is narrowing the gap, but the larger story is the rapid maturation of open-weight models,” Dai said. “Enterprises increasingly have credible alternatives to proprietary frontier models, particularly for software engineering, domain customization, sovereignty, and cost-sensitive deployments, where openness often matters as much as absolute model performance.” Company touts a 16-day autonomous coding run Alibaba said it tested the model on three unsupervised, multi-day coding projects requiring it to take a task from an empty project folder to completion without human assistance, including one project the company said took 16 days to complete on its own. Alibaba also highlighted enterprise applications across legal compliance, financial analysis, engineering design, quantitative research and multimodal content creation, saying the model is intended to complete entire business workflows rather than individual AI-assisted tasks. Amit Jena, development manager for AI at Kanerika, said that the claim deserves more scrutiny than it has received. “The claim worth examining is not the parameter count. Alibaba says the model completed a software engineering project in 16 days. That sentence has been reprinted everywhere and interrogated nowhere,” Jena said. “Sixteen days of what? How many times did a human step in? Did the output survive code review?” Jena said the open-weight commitment itself should also be read carefully. “Publishing weights is a separate act from opening an API endpoint,” he said. “Until there is a repository, a licence and a model card, open-weight describes an intention.” Analysts say inference efficiency isn’t the real constraint Alibaba’s mixture-of-experts architecture activates roughly 95 billion of the model’s 2.4 trillion parameters per request, a design the company says lowers inference costs. Dai said that tradeoff now matters more to enterprise buyers than raw model size. “Inference efficiency now matters more than raw model size for most enterprises,” he said. “Activating only a fraction of total parameters can significantly reduce serving costs and infrastructure requirements, making frontier-class performance more accessible for production deployments where scalability, latency, and economics are often bigger concerns than benchmark leadership.” Jena said efficiency gains matter less than an organization’s ability to actually test the model. “Efficiency stopped being the interesting question. The constraint that actually binds is evaluation throughput,” he said. Nitish Tyagi, senior principal analyst at Gartner, said the significance of the release lies less in the parameter count than in what it signals about competitive pressure on AI deployment costs. “Gartner has previously predicted that, without stronger cost controls, AI coding expenses could exceed the average developer’s salary,” Tyagi said. “The combination of open weights, a mixture-of-experts architecture, and a one-million-token context window represents a meaningful step toward making AI-augmented software development more economically viable.” Tyagi cautioned that enterprises need to look beyond inference costs when weighing the model for production use. “Many organizations outside China may be hesitant to rely on models hosted within China, leading them to deploy through hyperscalers or on-premises infrastructure, both of which introduce additional costs,” he said. Open-weight models also typically lack the indemnification protections that come with commercial AI vendors, he said, meaning enterprises need their own security, governance, and code-scanning controls to catch copyright and intellectual property risks before production deployment. What CIOs should look out for Jena said the flagship model announced Monday may not be the one enterprises end up running. “Qwen3.8-27B, announced alongside the flagship and almost entirely ignored in coverage,” is the more deployable option for most organizations, he said, since it can run on infrastructure they own and fine-tune on their own data. Dai said enterprise leaders evaluating the release should prioritize transparency and total cost of ownership over headline figures. “The key question is whether Qwen3.8 delivers measurable business outcomes, enterprise-grade reliability, lower total cost of ownership, and options for digital sovereignty compared with competing models,” he said. The article originally appeared on InfoWorld.