DeepSeek rolled out an updated model, V4.1-Flash-1017, with reported improvements in speed, response times, and benchmark scores.
On October 17, DeepSeek released V4.1-Flash-1017, updating its September V4.1-Flash model across its website, app and API. The company described it as a “minor update” that improves reasoning, coding, agentic work and visual benchmarks while shortening task time and output length. In DeepSeek’s reported tests, the model scored 77.0 on DeepSWE v1.1, up from 74.2; its 42.3 on Terminal-Bench 4.0 compared with 31.2 for the previous version. DeepSeek also reported 42.6 on Humanity’s Last Exam, versus 36.8. These are company-published results, not independent evaluations. The multimodal mixture-of-experts model has 552 billion total parameters, activating 8 billion during prefill and 16 billion during decoding. API pricing and access via the existing `deepseek-flash` identifier remain unchanged; reported off-peak rates are $0.15 per million uncached input tokens and $0.60 per million output tokens.