DeepSeek V4.1 Flash hands-on tests and comparisons

A user reports running DeepSeek V4.1 Flash at Max setting almost nonstop for two days in OpenCode Go and consuming only 11% of the monthly quota.

The poster shows DeepSeek-V4.1-Flash-1017 generating multiple simulations in one shot via direct API calls, with no harness, tools, or external assets, as a raw-capability demo.

A 748B model in Maya-S quant ran on a single 5090 via RAM/NVMe offload, finishing a voxel-world build in 74.1 min with 100,575 tokens written, 8.5/10 browser check and 22.9 tok/s code decode.

The author calculates $20 buys ~2.9B tokens on Claude Pro vs ~1.8B on DeepSeek V4.1 Flash because ~96% of billing is cache, yet at API list price those tokens cost ~$1,650 vs ~$33.

The author estimates ~35 tok/s and ~1.1B tokens/year locally on a 256GB M5 Ultra, costing $394 in electricity versus ~$663/year for equivalent off-peak API output, concluding the laptop is overspend if bought only for local LLMs.

A user complains V4.1 Flash got slower, yet measured throughput is still 150 tok/s on average and up to 169 tok/s, which the poster frames as rising expectations.

A tester lists 1017 agent flaws: budget counting inside thinking breaks, 'check status first' prompts fail, ambiguity handling is poor, and constraining it triggers a power-saving mode; coding remains strong, hitting 2K 60fps on a 3080Ti.

Strata added GSQ-RCO quantization support so DeepSeek V4.1-Flash can run locally on PCs with 64GB RAM and a 12GB+ GPU; speed is still limited but data stays on-device.

Strata testing reports IQ2_XXS quantization raises perplexity 15.5% over the base model, runs about 8–10 tok/s on an RTX 3060 12GB with 46GB RAM, and llama.cpp does not support it yet.

A user reports Strata now supports DeepSeek-V4.1-Flash-1017, letting 12GB VRAM plus 64GB DDR5 run all GSQ-RSO quants with Engram on SSD, with Maya comparisons coming.

An early comparison finds Qwen3.8-Flash-Next fast but prone to confident errors with long answers, while V4.1 Flash is slower yet cautious and concise, with little difference on simple tasks.

On an i7-14700 + RTX 5090 with 128GB RAM, the IQ2_XXS quant at 128K context and reasoning effort off ran at ~14 tok/s and finished a Minecraft-style 3D sandbox game prompt in 2h01m; quality is modest but a 763B model running on one GPU is the point.