VecSzn/PolyStrata

Maintainer? Show your league in your README

Card: your league's emblem, from Copper to Legendary

[![bestofgithub](https://bestofgithub.dev/badge/VecSzn/PolyStrata.svg?style=card)](https://bestofgithub.dev/repo/VecSzn/PolyStrata)

Flat: fits beside your other badges

[![bestofgithub](https://bestofgithub.dev/badge/VecSzn/PolyStrata.svg)](https://bestofgithub.dev/repo/VecSzn/PolyStrata)

Loading README…

Model	Laptop with an 8 GB card writes / reads	RTX 5090 writes / reads	Every size
LFM2.5-24B-A2B	70-73 / 3,590-3,620	480-485 / 23,390-23,460	LFM2.5
Flint 40B-A2.5B	61-63 / 2,000-2,010	251-252 / 12,900	Flint
Qwen3.6-35B-A3B	94-100 / 1,690	394-395 / 7,800-7,810	Qwen3.6
Clover-35B-A3B-Preview	68-73 / 1,560-1,700	not measured	Clover
Ornith 1.5	76-82 / 1,620	316-319 / 8,040-8,110	Ornith
Gemma 4 26B-A4B	92-93 / 2,240	353-356 / 8,420-8,480	Gemma 4
Tiel Coder 35B-A3B	83 / 1,700	337-345 / 8,080-8,110	Tiel Coder
MiniCPM-V 4.7 35B-A3B	64 / 1,150	308-311 / 7,910-7,950	MiniCPM-V
Kolibri 1	not measured	189-192 / 4,980-5,070	Kolibri 1
GLM 5.3 Flash	does not run: needs a 32 GB card	59-66 / 820	GLM
DeepSeek V4 Flash	does not run: needs a 32 GB card	56 / 1,040	DeepSeek

Model	Graphics card	RAM (the installer's size)	Disk	System
Qwen3.8-Flash-Next	NVIDIA RTX 20 to 50 series, or AMD RX 6800 / 6900, RX 7700 XT / 7800 XT / 7900 XT / XTX, RX 9060 XT / 9070 / 9070 XT, Radeon AI PRO R9700; 12 GB or more	32 GB or more; 64 GB runs every size	about 80 GB	Windows, Linux
GLM 5.3 Flash	NVIDIA RTX 20 series or newer; only run on a 32 GB card	96 GB beside a 32 GB card	79 to 114 GB, on an SSD	Linux
DeepSeek V4 Flash	same as above	72 GB beside a 32 GB card	91 GB, on an SSD	Linux
LFM2.5-24B-A2B	NVIDIA RTX 20 series or newer, 8 GB or more	about 20 GB beside an 8 GB card	11 to 25 GB	Windows, Linux
Flint 40B-A2.5B	same as above	about 22 GB beside an 8 GB card	18 to 42 GB	Windows, Linux
Qwen3.6-35B-A3B	same as above; AMD as a beta	about 21 GB beside an 8 GB card	13 to 23 GB	Windows, Linux
Clover-35B-A3B-Preview	NVIDIA RTX 20 series or newer, 8 GB or more	about 20 GB beside an 8 GB card	13 to 21 GB	Windows, Linux
Ornith 1.5	same as above; AMD as a beta	about 22 GB beside an 8 GB card	14 to 24 GB	Windows, Linux
Gemma 4 26B-A4B	same as above; AMD as a beta	about 18 GB beside an 8 GB card	11 to 17 GB	Windows, Linux
Tiel Coder 35B-A3B	same as above; AMD as a beta	about 21 GB beside an 8 GB card	14 to 23 GB	Windows, Linux
MiniCPM-V 4.7 35B-A3B	same as above; AMD as a beta	about 25 GB beside an 8 GB card	16 to 29 GB	Windows, Linux
Kolibri 1	NVIDIA RTX 20 series or newer, 8 GB or more	about 33 GB beside an 8 GB card	29 to 48 GB	Windows, Linux

Model	The installer takes	With less RAM	By name
Qwen3.8-Flash-Next	32 GB of RAM: Coder; 48 to 64 GB: IQ2_XS; 96 GB or more: IQ3_S		Q2_0, IQ3_XXS, Unsloth's 4- to 8-bit files
GLM 5.3 Flash	3.0BIT	UD-IQ2_XXS, then REAP50-Q3_K_M
DeepSeek V4 Flash	UD-IQ2_XXS (the only size)
LFM2.5-24B-A2B	UD-Q4_K_XL	UD-Q3_K_XL	QAD-Q4_0, UD-Q6_K_XL, Q8_0
Flint 40B-A2.5B	UD-Q4_K_XL from about 40 GB of RAM	UD-Q3_K_XL	UD-Q6_K_XL, Q8_0
Qwen3.6-35B-A3B	UD-Q3_K_M	UD-Q2_K_XL	UD-Q4_K_M
Clover-35B-A3B-Preview	Q3_K_M	APEX-I-NANOPLUS	Q4_K_M
Ornith 1.5	APEX-I-COMPACT	APEX-I-MINI	APEX-I-QUALITY
Gemma 4 26B-A4B	UD-Q3_K_M	UD-Q2_K_XL	UD-Q4_K_M, QAT-UD-Q4_K_XL
Tiel Coder 35B-A3B	UD-Q3_K_XL	UD-IQ3_XXS	UD-Q4_K_XL
MiniCPM-V 4.7 35B-A3B	Q4_K_M	mix_q3gud_im	Q6_K
Kolibri 1	Q4_K_M from about 48 GB of RAM	Q2_K	Q3_K_M

	Qwen3.8-Flash-Next	GLM 5.3 Flash	DeepSeek V4 Flash	LFM2.5-24B-A2B	Flint 40B-A2.5B	Kolibri 1	Qwen3.6-35B-A3B, Clover-35B-A3B-Preview, Ornith 1.5, Gemma 4 26B-A4B, Tiel Coder 35B-A3B, MiniCPM-V 4.7 35B-A3B
Pictures	Say yes to "Images?" in setup. Then click Picture in the chat, or attach pictures in your app. AMD cards read pictures on Linux through the processor; on Windows they can't yet.	Not yet.	It reads text only.	It reads text only.	It reads text only.	It reads text only.	Say yes to "Images?" in setup, then as with Qwen3.8-Flash-Next. On a card with less than 16 GB the picture encoder runs on the processor, and a picture takes a few seconds. Clover reads text only.
Several requests	One at a time by default, the others wait. `"parallel": 2` answers several at once (BATCHING.md); on a 12 GB card each answer gets slower.	One at a time, the others wait.	One at a time, the others wait.	One at a time, the others wait.	One at a time, the others wait.	One at a time, the others wait.	One at a time, the others wait.
Several conversations	The last one is kept, so follow-up messages start in seconds. Two engine arguments keep several in RAM (details).	The last one is kept. Setup asks whether to keep several in RAM, so that an agent and its subagents take turns without their prompts being read again (how).	As GLM 5.3 Flash.	As GLM 5.3 Flash.	As GLM 5.3 Flash.	As GLM 5.3 Flash.	As GLM 5.3 Flash.
Long prompts	The first message of a chat is read in full, about 1 minute per 30,000 tokens.	The first message is read in full, about 40 seconds per 30,000 tokens.	The first message is read in full, about 29 seconds per 30,000 tokens.	The first message is read in full: 30,000 tokens take 1.3 to 1.5 seconds on the workstation and 8 to 10 seconds on the 8 GB laptop.	The first message is read in full: 30,000 tokens take about 2.3 seconds on the workstation and 15 to 22 seconds on the 8 GB laptop.	The first message is read in full: 30,000 tokens take about 6 seconds on the workstation with Q2_K.	The first message is read in full: 30,000 tokens take 4 to 5 seconds on the workstation and 14 to 27 seconds on the 8 GB laptop.

VecSzn/PolyStrata

PolyStrata

How fast is it?

What you need

Install

Let your AI set it up

Or do it yourself

Which model should I pick?

Using it

Optional: more speed on one card when the whole model fits in RAM

Something went wrong?

How does it work?

Credits and license