Alibaba's Qwen3.8-Max Hits 2.4T Params, Tops Fable 5
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Alibaba released Qwen3.8-Max, a 2.4T-parameter AI model, claiming it tops Kimi K3 on some benchmarks alongside performance comparable to Anthropic's Fable
- Alibaba plans to release open weights for Qwen3.8-Max and Qwen3.8-27B next week, with the 27B variant expected to run locally on 17GB RAM/VRAM setups
- Qwen3.8-Max scored 2% higher than Anthropic's Fable 5 on Terminal-Bench benchmarks, per cline's comparison
- The release comes just days after DeepSeek V4-Flash claimed similar performance, suggesting open-weight models have now surpassed closed models
- Some practitioners — including Chahat Sharma — cautioned that the final license terms remain unconfirmed until weights actually drop, noting Qwen's open releases have historically used Apache 2.0 but a Max-class model is where usage restrictions typically appear
- Unsloth said the 27B version will run on 17GB VRAM setups, putting flagship-capable performance within reach of consumer hardware
Why it matters: Qwen3.8-Max's 2% edge over Anthropic's Fable 5 on Terminal-Bench — achieved just days after DeepSeek V4-Flash's similar claim — shows Chinese open-weight models now reliably match closed frontier models, while the 27B variant running on 17GB of VRAM puts near-frontier capability onto a single consumer GPU.
