DeepSeek V4.1 Flash hands-on tests and comparisons

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.

Austria's DASLab quantized the 763B-parameter MoE to 2.00 bpw (138.2GB), claims it matches the base model on AIME25 and GPQA-Diamond, and runs it on one 12GB GPU via a memory-mapped 110.6GB Engram file using a custom Strata runtime.

Report of running DeepSeek V4.1-Flash with GSQ-RCO quantization and the Strata engine on 12GB VRAM plus 64GB RAM at 105–117 tok/s, with a caveat that prefill and decode speeds differ.

On the same model, basic Git tasks like listing the last 10 commits took about 10 seconds in Qoder versus 20–30 seconds in DeepSeek Harness.

A user reports that locally hosted stock (non-abliterated) DeepSeek V4.1 1017 weights, with a simple system prompt, readily took on a request to find RCE or root access on a Samsung TV.

A user ranks seven budget models, placing DeepSeek V4.1 Flash 1017 first for speed, price and intelligence, especially with the DSH harness; GLM 5.3 Flash is second but slow and prone to cutting off.

A user says WorkBuddy is the best low-cost route to DeepSeek V4.1 Flash 1017 at 0.11x credits (and Kimi K3.1 Pro at 0.97x), with free credits usable.

A promotional post says DeepSeek V4.1 Flash 1017 is available in OpenCode, Claude Code, Pi and omp, scoring 77.0 on DeepSWE versus Opus 5's 74.0, while still trailing on Terminal-Bench at 42.3 vs 51.8.

A user reports a clear experience gap between the free DeepSeek V4.1 Flash in OpenCode and the official API, suspecting they are not the same model, without concrete tests.

Personal-plan users can pick V4.1-Flash-1017 and run thinking at maximum effort without spending credits; the poster suggests using the week to save Claude and Codex quota.

A user ran head-to-head real-world agent tasks between Ornith 1.5 122B EXL3 on four Intel B70s and DeepSeek V4.1 Flash 1017 EXL3 on two Sparks.

A four-model volcano rendering comparison (Kimi K3.1 Pro, GPT-5 Astra Max, DeepSeek-V4.1-Flash Max, GLM-5.3-Flash Max) finds Kimi's 3D modeling notably more detailed, with per-model costs listed.