DeepSeek V4.1 Flash hands-on tests and comparisons

A user notes that DeepSeek V4.1 Flash 1017 is now available in Cursor and starts using it for coding, with a screenshot of the setup.

DeepSeek V4.1 Flash is now available in Cursor, costing $0.60–$0.98 per task on CursorBench, making it a budget-friendly option for everyday coding.

User runs V4.1 Flash 1017 in MXFP4 quantized form, pipeline-split across a MacBook Pro M5 Max and Mac Studio M5 Ultra, and plans to try the Muse Spark open release next.

A roundup of a small survey comparing DeepSeek V4.1 Flash coding experience in Codex, Claude Code, Pi and the official DSH harness, arguing the harness shapes prompt, context compaction and tool use.

A user asks for quantization and inference recipes to run DeepSeek V4.1 Flash 1017 locally on an M5 Ultra with 256GB of memory.

The author says DeepSeek 1017 is cheaper and better than Grok 4.7 on the same CursorBench, and questions Cursor subscriptions that no longer include non-Grok models.

A developer building a local music app found that DeepSeek V4.1 Flash plays Radiohead's Creep every single time it tests a feature.

Running DeepSeek-V4.1-Flash-1017 locally on an M5 Ultra 256GB with an MLX recipe, hitting 46.3 tok/s on long 1M-context text and 50 tok/s on normal prose.

User runs V4.1 1017 at 3.75bpw on a custom TP3 DGX Spark cluster and one-shots a household banking app that parses PDF statements into a SQL ledger, calling the local model result impressive.

The author argues the V4.1 Flash 1017 build leads OpenRouter's paid developer traffic on price and token efficiency, not overall capability.

A user notes DeepSeek V4.1 Flash 1017 is already on Modal, letting people try the new open model the week it launches without their own GPUs.

Item 8 of a daily digest notes Cursor announced DeepSeek V4.1 Flash and reported its CursorBench cost-frontier results, cautioning that per-task pricing depends on test setup.