Back
Alibaba unveils 2.4 trillion-parameter Qwen3.8-Max, publishes benchmarks and open-weights plan

Alibaba unveils 2.4 trillion-parameter Qwen3.8-Max, publishes benchmarks and open-weights plan

Alibaba says Qwen3.8-Max will be released for free as the first open-weight Max-class Qwen, while benchmark data and prediction-market pricing suggest more pressure on Anthropic and other frontier-model rivals.

Fact Check
The core factual claims are confirmed by Alibaba's own press release and multiple independent outlets. The Alibaba Cloud press release confirms 2.4 trillion parameters and that weights will be released 'next week,' matching the claim. CNBC and SiliconANGLE confirm the parity/rivalry framing with Claude Fable 5 and the context of the Chinese AI race following Moonshot AI's Kimi K3. The only minor nuance is that Alibaba's self-described claim is 'second only to Fable 5' rather than strict 'parity'; however, CNBC characterizes performance as 'competitive with' and 'rivaling' Fable 5, so the headline's 'claims parity' is a reasonable, non-misleading paraphrase.
Summary

Alibaba launched Qwen3.8-Max, a 2.4 trillion-parameter model it says will be released for free as open weights, marking the first time a Max-class Qwen model will be distributed that way. Alibaba presented the system as nearing the capabilities of Claude- and ChatGPT-class tools, while its own benchmark table showed a more mixed picture: Anthropic’s Fable 5 led 15 of 31 text tests, OpenAI’s GPT-5.6 Sol led nine, and Qwen3.8-Max led seven, with Qwen performing more strongly on multimodal work than on core coding evaluations. Alibaba said the model activates 95 billion parameters at a time, supports a 1 million-token context window and can work inside Anthropic’s Claude Code and OpenAI’s Codex. The company highlighted autonomous runs including 16 days spent building a coding tool, and it previously said open weights for Qwen3.8-Max and Qwen3.8-27B would be posted on Hugging Face and ModelScope. Prediction-market pricing shifted slightly after the announcement, suggesting traders see lower odds that Anthropic will still hold the top AI model by the end of September 2026. Markets are now watching adoption, benchmark results and any response from Anthropic or other major developers.

Terms & Concepts
  • Open weights: A model release approach that lets users download the trained parameters for local use or further development.
  • Parameters: The internal numerical settings of an AI model that shape how it processes prompts and generates outputs.
  • Prediction-market pricing: Market prices used to infer how traders collectively assess the probability of a future outcome.