"Wishful" coding capabilities upgraded: DeepSeek launches new self-developed search API, just 5 yuan for a thousand calls

"Something big is coming." DeepSeek, the leading domestic large model developer, suddenly went down a few days ago, sparking speculation among many AI enthusiasts that it was about to release a major update—and sure enough, that's exactly what happened. According to The Paper reporter on October 17, DeepSeek has rolled out another version update: DeepSeekV4.1 Flash has been upgraded to version V4.1-Flash-1017, with the official web interface, app, and API models all switched to the new model in sync. According to the official description, compared with the previous version DeepSeekV4.1 Flash, the new model has made substantial improvements across various benchmarks. At the same time, while boosting performance, the new model has shortened task runtime and output length, enabling users to complete tasks at lower cost and faster speed, and achieving stronger cost-effectiveness across multiple Harnesses (officially called