AIDeepSeek and Alibaba Release Efficient Multimodal Models with Compression Advances
DeepSeek released V4.1-Flash (552B MoE with ~8B/16B active params) emphasizing KV cache compression and multimodal support, while Alibaba released Qwen3.8-Omni-Flash with omni-modal capabilities and 1M-token context.
TLDR
Chinese labs continue rapid iteration on efficient, capable models at lower costs, pressuring the ecosystem and driving API price expectations downward. Multimodal and long-context strengths appeal to builders, fueling discussions around compute competition and accessibility of advanced models.
Combined views
—
2 Sources, first seen 6h ago
DeepSeek and Alibaba Release Efficient Multimodal Models with Compression Advances
DeepSeek released V4.1-Flash (552B MoE with ~8B/16B active params) emphasizing KV cache compression and multimodal support, while Alibaba released Qwen3.8-Omni-Flash with omni-modal capabilities and 1M-token context.
TLDR
Chinese labs continue rapid iteration on efficient, capable models at lower costs, pressuring the ecosystem and driving API price expectations downward. Multimodal and long-context strengths appeal to builders, fueling discussions around compute competition and accessibility of advanced models.