DeepSeek Open Source V4.1-Flash Update: Outperforms Claude Opus 5 in Multiple Programming and Agent Tests DeepSeek has officially released V4.1-Flash-1017, a minor update to V4.1-Flash. It's now live on the official website, app, and API, with model weights open-sourced on Hugging Face under the MIT license. The model boasts a parameter size of 763 billion and supports up to 1 million tokens in context. According to the official announcement, the new version delivers better test results, faster task completion, and shorter outputs. On the programming test DeepSWE, it scored 77.0, surpassing Claude Opus 5's 74.0 and GPT-5.6 Sol's 73.0. It also ranked first in several agent tests like Automation-Bench and Agents’ Last Exam. The cost is even more impressive: on DeepSWE, it averages just $0.2 per task, while Claude Opus 5 costs over $10 per task. However, it still lags behind Claude Opus 5 in tests like HLE and Terminal-Bench.