DeepSeek V4.1 Flash hands-on tests and comparisons

OpenRouter data for Oct 19–25 shows V4.1-Flash-1017 first at 33.6T tokens, up 801% week over week, with Chinese models taking all top five spots.

On Oct 17 DeepSeek released V4.1-Flash-1017 across web, app and API at unchanged pricing, published MIT-licensed weights, and kept the Harness agent tool in developer preview.

A 763B MoE compressed for local runs via non-uniform quantization; the GSQ-RCO build balances size and accuracy and ships as GGUF.

IST DASLab's per-tensor gradient-searched quantization of the 763B MoE comes in 1.75/2.00/2.25 bit tiers, runs on a 12GB card with Strata, is MIT-licensed, and has passed 410k downloads.

A user recaps DeepSeek's releases from April to October: V4 preview, a permanent 75% price cut, DSpark and DeepSpec, V4-Flash GA, V4-pro-0813, V4.1-flash and DSec, ending with V4.1-flash-1017 and DeepInfer in October.

A Japanese account summarizes V4.1-Flash-1017: big gains on DeepSWE v1.1 and other benchmarks, self-feedback RL enabling autonomous 3D and game projects, fewer tokens and less time per task, and MIT-licensed open weights.

A daily news roundup notes IST Austria researchers made a quantized version of the 763-billion-parameter DeepSeek V4.1 Flash that runs locally on a 12 GB VRAM card.

In an AI news roundup, DeepSeek-V4.1-Flash-1017 at max setting scores 57% on Artificial Analysis' defense-only Cyber Index, ranking first among publicly available models.

The release swaps in new weights across web, app and API; DeepSeek claims shorter outputs, faster tasks and higher coding/agent scores in its own tests, with Harness v0.2.1-rc.1, open Search/Fetch APIs and HuggingFace weights.

The model card lists Terminal-Bench 4.0 rising from 31.2 on the previous Flash to 42.3, plus deepseek-recipe Rust libraries with Python bindings.

An overview of the October 17 V4.1 Flash '1017' release: MIT-licensed open weights, 1M-token context, and a vendor-reported DeepSWE coding score of 77.0, runnable on private servers.

A roundup says DeepSeek shipped V4.1-Flash-1017 weights under MIT, open-sourced its DeepInfer serving engine, and is reportedly raising about $15B.