DeepSeek released the V4.1-Flash-1017 model on October 17, maintaining previous pricing. The model weights are available under the MIT license, and updates are live across web, app, and API platforms.
In comparative programming tests, DeepSeek V4.1-Flash-1017 completed tasks at an average cost of $0.045 per task, compared to $0.1168 for Claude Opus 5.5 and $0.26 for GPT-6 Astra.
While cost-efficient, DeepSeek's processing time averaged 2 minutes and 38 seconds per task, significantly slower than the 37 seconds for Opus and 55 seconds for GPT-6 Astra.
Performance varies by environment and settings. When switching to a lower inference tier, DeepSeek's time dropped to 1 minute and 31 seconds at a cost of $0.0257, though success rates fluctuated.
DeepSeek's efficiency is supported by a 97.9% cache hit rate. However, competitors like GPT-6 Luna and Haiku 5.5 also offer competitive pricing, with costs as low as $0.0145 per task in similar tests.
