DeepSeek V4.1 Flash hands-on tests and comparisons

In Cursor, DeepSeek V4.1 Flash scores 47.7% for $0.98 versus Grok 4.7 Extra High's 46.3% for $6, six times the price.

The author compares GPT-6.1 Terra, Claude Sonnet 5.5, Haiku 5.5 and DeepSeek V4.1 Flash on price, intelligence and speed, finding V4.1 Flash cheapest and GPT-6.1 Terra at only 141 tok/s even with fast mode.

DeepSeek V4.1 Flash landed in Cursor, hitting 47.7% on CursorBench at max effort for under a dollar per task versus roughly 57% for Opus 5.5.

A user ran the same pelican-riding-a-bicycle prompt on GPT-6.1 Terra Max and DeepSeek V4.1 Flash and found the two roughly on par.

Per a Cursor chart, DeepSeek V4.1 Flash's API price-performance sits roughly between Haiku extra high and Sonnet high, though subscription discounts change the picture.

Tuning Strata on an RTX 5090 with a 2.25bpw 32GB quantization of V4.1 Flash yields ~50–51 tok/s long generation at 256K context and 1350 tok/s cold prefill.

A home setup with a 4090D 48GB and DDR4 runs the 763B-parameter DeepSeek V4.1 1017 build at about 30 tok/s decode, buffering the full 1M context for roughly 2k tok/s prefill.

Links to the ISTA-DASLab DeepSeek-V4.1-Flash-1017 GSQ-RCO GGUF model card with per-tensor allocation table, Strata's run docs, and the GSQ and RCO papers.

The author applied an Adaptive Experts technique to a vLLM deployment of DeepSeek Flash 1017 on a DGX Station GB300, reporting a solid speedup without a sidecar and expecting it to transfer to other MoE models, with a GitHub recipe linked.

On the 226-problem open math benchmark it places 6th, tying Gemini 4 Argon and Muse Wonder at much lower cost, prompting questions about closed-model premiums.

The SGLang team says MegaKernel lifts interactivity to 943 tok/s per user, while WideEP and system optimizations raise total throughput per GPU from 348K to 654K tok/s.

Artificial Analysis shows sample submissions for one brief with cost and output tokens per task: Opus 5.5 at 1536, DeepSeek V4.1 Flash at 1053, MiniMax-M3 at 724, Nemotron 3 Ultra at 304.