Qwen 3.8-Flash-Next From Alibaba: The First Look at Qwen 4
The 125-billion-parameter model uses only 6 billion active parameters per token and is available as open weights for local deployment.
4 Articles
4 Articles
Qwen3.8-Flash Matches DeepSeek V4 Pro at a Quarter of the Price
Alibaba’s Qwen team released Qwen3.8-Flash on August 26, 2026, a multimodal mixture-of-experts model that activates only 6 billion parameters per token despite carrying 125 billion total. On paper, it is an efficiency play. In practice, it beats DeepSeek V4 Pro on SWE-bench Pro (62.5 vs. 55.4) while costing roughly a quarter of the price, and it lands within striking distance of Claude Opus 5 on coding benchmarks at 33x lower input cost. The rel…
Qwen 3.8-Flash-Next From Alibaba: The First Look at Qwen 4
Alibaba’s Qwen team didn’t wait for a big stage. On August 26, 2026, they quietly released Qwen 3.8-Flash-Next — a 125 billion parameter AI model that’s designed to run on hardware most developers already own. It’s not the finished Qwen 4. It’s the architecture preview, the sneak peek that tells developers what’s coming and lets them start building now. Here’s what’s inside, what’s missing, and why it matters. What Qwen 3.8-Flash-Next Actually I…
Alibaba’s QwenWork adds Qwen3.8-Flash with a new Standard mode · TechNode
Alibaba’s QwenWork has added Qwen3.8-Flash and launched a new Standard mode that is available to all users. The service will now offer Standard and Advanced modes for different levels of office tasks. Qwen says about 95% of daily tasks can be handled by the Standard mode, while more complex tasks require the Advanced mode. In tests cited by the company, the new mode increased single-task generation speed by about 100% and reduced token consumpti…
Alibaba continues to accelerate on open models with Qwen3.8-Flash-Next, a new multimodal model Mixture-of-Experts thought to be an advanced overview of architecture that will serve as a basis for Qwen4. On paper, the recipe is ambitious: 125 billion parameters for the main model, 51 billion more dedicated to N-gram Embeddings, but only 6 billion [...] L的article Qwen3.8-Flash-Next: 125 billion parameters, 1 million tokens and an overview of Qwen4…
Coverage Details
Bias Distribution
- 67% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
