Alibaba and DeepSeek release Flash models with reported 1M-token context windows
TestingCatalog reports that Alibaba’s Qwen3.8-Omni-Flash understands audio and video. A separate account says DeepSeek V4.1 Flash is live with native vision and shares claims from its own lab tests.
TLDR
TestingCatalog reports that Alibaba released Qwen3.8-Omni-Flash with audio and video understanding and a 1M-token context window. It describes performance as close to Gemini 3.8 Flash on WildClawBench-MM and UniClawBench. Separately, an account sharing its own lab-test results says DeepSeek V4.1 Flash matched GPT-5.6 Sol and beat Claude Opus 5 on agentic benchmarks. That account also describes native vision and a 1M-token context window, and says the model is available as deepseek-flash.
Combined views
—
2 Sources, first seen ago
