Overview
- DeepSeek announced V4.1-Flash-1017 on Oct. 17, 2026, saying the model is faster, produces shorter outputs and posts stronger benchmark scores than its prior Flash release.
- The company published technical notes and open-sourced the model weights under an MIT license on Hugging Face and ModelScope to lower barriers for developers.
- DeepSeek reported a 77.0% score on the DeepSWE v1.1 coding test and HLE results of 36.8–42.6, figures the firm says reflect improved reasoning, coding and agentic abilities.
- For the first time DeepSeek opened its Search & Fetch APIs to outside developers with introductory free access until Nov. 1 and pricing of $0.75 per 1,000 calls.
- Operational risks persist after a recent roughly five-hour outage and growing slowdowns, and independent testing has not yet verified DeepSeek’s performance and reliability claims.