DeepSeek launches V4.1-Flash beta with enhanced visual intelligence and performance

On September 8, 2026, DeepSeek announced the start of internal testing for its V4.1-Flash interim beta version. This update features a new model architecture, enabling native multi-modal support while improving overall performance, speed, and cost efficiency. Developers can access the model by setting the model name to 'deepseek-v4.1-flash-expires-on-0910' without changing the base URL, with billing rates remaining consistent with the previous version. Earlier this year, DeepSeek introduced V4-Flash-Vision-Exp, an experimental multi-modal model that outperformed its predecessor in visual understanding benchmarks, nearing the capabilities of Opus-4.8. The company also released DeepSeek Harness v0.1 on August 13, which quickly gained traction, receiving over 70,000 GitHub stars within 24 hours. Following the launch of V4 Pro, DeepSeek adjusted its API pricing system to include peak-valley time-based pricing. The internal testing of V4.1-Flash marks a significant step in its multi-modal and performance optimization efforts.