Alibaba has launched Qwen 3.8-Max, its largest artificial intelligence model to date. The company previewed the model on July 19 at the World AI Conference in Shanghai and released it publicly on August 3, 2026 with full API access.
What Makes Qwen 3.8-Max Different
The model packs 2.4 trillion parameters in a sparse Mixture-of-Experts architecture, with roughly 95 billion active parameters per token. That puts it as the second-largest publicly known AI model behind Moonshot’s Kimi K3 at 2.8 trillion parameters, which launched the same week.
Alibaba’s own claim is that Qwen 3.8-Max is “second only to Claude Fable 5” among frontier models. No benchmark table has been published yet, so this ranking rests entirely on internal evaluations.
The model supports text and visual inputs, with a 1 million token context window and up to 128k output tokens. API pricing stands at $2 per million input tokens and $6 per million output, undercutting competitors like Kimi K3’s $3 in / $15 out.
Open Weights Promise
Unlike previous Max-tier Qwen models which stayed API-only, Alibaba has committed to releasing open weights for both Qwen 3.8-Max and a smaller Qwen 3.8-27B variant within days of launch. As of this writing, neither has appeared on Hugging Face.
For developers who need open weights immediately, the practical options remain Qwen 3.6, GLM 5.2, and Kimi K3.
How It Compares
The active parameter count of 95B per token puts Qwen 3.8-Max in a lighter serving class than Kimi K3’s 104B active parameters. For production teams, this translates to lower per-token cost and potentially faster inference.
The model supports both OpenAI and Anthropic API specifications, making it straightforward to integrate into existing codebases. Alibaba’s previous Qwen 3.7-Max posted strong results on coding benchmarks, including 80.4 on SWE-bench Verified and 92.4 on GPQA Diamond.
If 3.8-Max improves on those numbers while adding multimodal support, the competitive positioning claim becomes plausible. Independent benchmarks should appear in the coming weeks.
Who Should Use It
Alibaba positions Qwen 3.8-Max for coding, full-stack development, data analysis, and office workflows. The multimodal capabilities expand its use cases beyond text-based tasks.
Indian developers and startups can access the model through Alibaba’s DashScope API with regions in Singapore and the US. The pricing makes it one of the more affordable frontier options compared to OpenAI and Anthropic’s offerings.
FAQ
When was Qwen 3.8-Max released?
Alibaba previewed it on July 19, 2026 at the World AI Conference in Shanghai and launched it publicly on August 3, 2026 with standard API access.
Is Qwen 3.8-Max open source?
Not yet. Alibaba promised to release open weights within days of launch for both the flagship and a smaller 27B variant. Neither has appeared on Hugging Face as of this writing.
How much does Qwen 3.8-Max cost?
The standard API pricing is $2 per million input tokens, $6 per million output tokens, and $0.25 for cached input tokens.
Is Qwen 3.8-Max better than GPT-5?
Alibaba claims it trails only Claude Fable 5 among frontier models. No independent benchmark comparison with GPT-5 has been published yet.
Can I run Qwen 3.8-Max locally?
Not yet. The model requires significant GPU resources at 2.4 trillion parameters. Open weights have not been released. For self-hosting, the smaller Qwen 3.6 models remain available.
