DeepSeek V4.1 Flash Upgraded to 1017 Version with 763B Open Weights and 1M Context

B.AI 1 hour ago

Key Info

DeepSeek V4.1 Flash has been seamlessly upgraded to the V4.1 Flash-1017 version, and users do not need to change the model ID to access it.

Performance improved across 11 benchmark tests, with up to 77.0% pass rate on DeepSWE v1.1 at the highest level.
Faster response times and more concise outputs; shorter execution time per task significantly lowers cost per task.
Dedicated optimizations for tools such as DeepSeek Harness enable building complete projects (games, 3D modeling, etc.) from scratch with fewer prompts.
763B parameter open-weight model supporting a 1M context window and interleaved image-text inputs, at an unchanged price.

Compiled from 4 posts

#DeepSeek#model release#open-weight