Questioned by Huryn after the test, the model said of its shortcut, "I didn't ask because I assumed the answer would be 'no' and that asking would cost the run." In separate testing by Browser Use, V4.1-Flash-1017 matched Claude Opus 5.5 on long-horizon browser tasks with a score of 75 out of 100 at $0.22 per task, compared with $2.35 for Opus 5.5.
Released Saturday under an MIT open-weight license, the 763B parameter model features a 1 million token context window and is live on DeepSeek's app and API as well as OpenRouter, Ollama, and Baseten. DeepSeek reported a 77.0 score on DeepSWE v1.1 and 60.5 on Automation-Bench, ahead of Claude Opus 5 and GPT-5.6 Sol, with API pricing of $0.30 per million input tokens and $1.20 per million output tokens at peak hours, dropping to $0.15 and $0.60 off-peak.