Alibaba Open-Sources 2.4 Trillion Parameter Qwen3.8 Flagship Model
大模型之路 · wechat · 2026-08-14
Alibaba's Qwen team has released the weights for Qwen3.8-2.4T-A95B, marking the first time a Max-level flagship model has been open-sourced in China. The model features 2.4 trillion total parameters with 95 billion activated per token, natively supporting a 262,144-token context window extendable to over one million.
This move significantly lowers the barrier to entry for top-tier foundation models. Using 1-bit layered quantization, the model size is compressed to 397GB, allowing enterprises to run local private deployments on a single high-end workstation, thereby avoiding steep cloud API costs. On the same day, DeepSeek V4 Pro officially launched, while Elon Musk's Grok 4.6 remained closed-source via API, highlighting a direct clash between open and closed ecosystems in the long-horizon autonomous agent track.
Although the open-source version slightly trails top closed-source models on certain coding benchmarks and has updated commercial licensing, its low-cost, highly flexible deployment is putting real pressure on the pricing power of closed-source vendors. As domestic open-source models accelerate their iteration pace, the industry's competitive focus is shifting from raw model capabilities to ecosystem building and infrastructure adoption.
More from Infra
- MLCommons launches MLPerf Endpoints, a benchmark that measures inference endpoints the way buyers actually purchase them — TheKanter · 2026-10-16
- AI-Driven Timing-Aware Power Optimization Uses Learned Ranking With Baseline-Preserving Repair — Amit Sharma · 2026-10-16
- Nscale AI datacenter plagued by coolant leaks, transformer fire, 10 safety violations — saranormous · 2026-10-16
- WSL 3.0 ships with native containers and Linux 6.18 kernel, syscall overhead down ~16% — solyarisoftware · 2026-10-16
- Are 22 GB modded RTX 2080 Ti cards a viable budget local LLM option? — Dave86ch · 2026-10-16
- Intel's neural-compressor: SOTA low-bit LLM quantization with FP4/NVFP4 support — HaihaoShen · 2026-10-16