Models

Flint

RecursiveSI · October 2026

Get

Hugging Face

A 40B, 2.8B active chat model from RecursiveSI.

Downloads: 469K
Likes: 404
Parameters: 40B, 2.8B active
Context: 256K
Licence: apache-2.0

Tight fit on M4 · 16 GB

The IQ2_XXS size from bartowski is the one to get: a 12 GB download that uses 12 GB of memory, and writes about 19 words a second.

Ratings

Software fixesSWE-bench Verified: 74.2%

SWE-bench Verified: real GitHub issues fixed so the project's tests pass. Scores are the ones filed on the model's Hugging Face page or the board's own table, read 2026-11-05; some come from the maker.

Flint runs well from a Mac with an M3 Pro and 18 GB of memory. 48 of 59 Mac configurations run it well.

Compare with

The nearest models in size.

Sizes

From unsloth/flint-GGUF on Hugging Face. A smaller size is a smaller download and a little less accurate.

Size	Download	Memory at 8K	Get
UD-IQ1_M	11 GB	12 GBestimated	Open
UD-IQ2_XXS	12 GB	12 GBestimated	Open
UD-Q2_K_XL	14 GB	14 GBestimated	Open
UD-IQ3_XXS	15 GB	15 GBestimated	Open
UD-Q3_K_XL	18 GB	19 GBestimated	Open
UD-Q4_K_XL	25 GB	25 GBestimated	Open
UD-Q5_K_XL	30 GB	30 GBestimated	Open
UD-Q6_K_XL	36 GB	37 GBestimated	Open
Q8_0	42 GB	43 GBestimated	Open
BF16	80 GB	80 GBestimated	Open

Run It Anywhere

Flint is an open model, so any app that runs GGUF files can run it. These take the IQ2_XXS size from bartowski straight from Hugging Face.

LM StudioMac, Windows, LinuxOpen
JanMac, Windows, LinuxOpen
MstyMac, Windows, LinuxOpen
OllamaMac, Windows, Linuxollama run hf.co/bartowski/RecursiveSI_flint-GGUF:IQ2_XXS
llama.cppMac, Windows, Linuxllama-server -hf bartowski/RecursiveSI_flint-GGUF:IQ2_XXS
Docker Model RunnerMac, Windows, Linuxdocker model run hf.co/bartowski/RecursiveSI_flint-GGUF:IQ2_XXS

On a PC Graphics Card

Video memory	Largest size that fits	Download	Memory at 8K
6 GB	None worth running		needs 14 GB
8 GB	None worth running		needs 14 GB
10 GB	None worth running		needs 14 GB
11 GB	None worth running		needs 14 GB
12 GB	None worth running		needs 14 GB
16 GB	UD-IQ3_XXS	15 GB	15 GB
20 GB	UD-Q3_K_XL	18 GB	19 GB
24 GB	UD-Q3_K_XL	18 GB	19 GB
32 GB	UD-Q5_K_XL	30 GB	30 GB
48 GB	Q8_0	42 GB	43 GB

The largest size, of those that keep three bits a weight or more, whose memory at an 8K chat is under 92% of the card's; the rest is the desktop's. Estimated from the files, the same way as for a Mac; speed on a PC is not estimated yet. A size that doesn't fit the card still runs, slower, with part of it in ordinary memory.

On every Mac

Mac	Size	Fit	Memory	Writes
M6 · 16 GB	IQ2_XXS	Tight fit	12 GB	32
M6 · 24 GB	UD-IQ3_XXS	Runs well	16 GB	29
M6 · 32 GB	Q3_K_L	Runs well	21 GB	25
M5 · 16 GB	IQ2_XXS	Tight fit	12 GB	31
M5 · 24 GB	UD-IQ3_XXS	Runs well	16 GB	28
M5 · 32 GB	Q3_K_L	Runs well	21 GB	23
M5 Pro · 24 GB	UD-IQ3_XXS	Runs well	16 GB	50
M5 Pro · 48 GB	Q4_0	Runs great	23 GB	46
M5 Pro · 64 GB	UD-Q5_K_XL	Runs great	31 GB	35
M5 Max · 36 GB	Q4_0	Runs well	24 GB	63
M5 Max · 48 GB	Q4_0	Runs great	23 GB	79
M5 Max · 64 GB	UD-Q5_K_XL	Runs great	31 GB	60
M5 Max · 128 GB	Q8_0	Runs great	47 GB	58
M5 Ultra · 96 GB	Q8_0	Runs great	45 GB	96
M5 Ultra · 256 GB	Q8_0	Runs great	47 GB	96
M5 Ultra · 512 GB	Q8_0	Runs great	47 GB	96
M4 · 16 GB	IQ2_XXS	Tight fit	12 GB	19
M4 · 24 GB	UD-IQ3_XXS	Runs well	16 GB	18
M4 · 32 GB	Q3_K_L	Runs well	21 GB	16
M4 Pro · 24 GB	UD-IQ3_XXS	Runs well	16 GB	30
M4 Pro · 48 GB	Q4_0	Runs great	23 GB	29
M4 Pro · 64 GB	UD-Q5_K_XL	Runs great	31 GB	23
M4 Max · 36 GB	Q4_0	Runs well	24 GB	36
M4 Max · 48 GB	Q4_0	Runs great	23 GB	41
M4 Max · 64 GB	UD-Q5_K_XL	Runs great	31 GB	33
M4 Max · 96 GB	Q8_0	Runs great	45 GB	29
M4 Max · 128 GB	Q8_0	Runs great	47 GB	34
M3 · 8 GB	UD-IQ1_M	Too big	16 GB
M3 · 16 GB	IQ2_XXS	Tight fit	12 GB	18
M3 · 24 GB	UD-IQ3_XXS	Runs well	16 GB	16
M3 Pro · 18 GB	UD-IQ1_M	Runs well	12 GB	24
M3 Pro · 36 GB	Q4_0	Runs well	24 GB	21
M3 Max · 36 GB	Q4_0	Runs well	24 GB	33
M3 Max · 48 GB	Q4_0	Runs great	23 GB	39
M3 Max · 64 GB	UD-Q5_K_XL	Runs great	31 GB	31
M3 Max · 96 GB	Q8_0	Runs great	45 GB	25
M3 Max · 128 GB	Q8_0	Runs great	47 GB	30
M3 Ultra · 96 GB	Q8_0	Runs great	45 GB	44
M3 Ultra · 256 GB	Q8_0	Runs great	47 GB	44
M3 Ultra · 512 GB	Q8_0	Runs great	47 GB	44
M2 · 8 GB	UD-IQ1_M	Too big	16 GB
M2 · 16 GB	IQ2_XXS	Tight fit	12 GB	18
M2 · 24 GB	UD-IQ3_XXS	Runs well	16 GB	16
M2 Pro · 16 GB	IQ2_XXS	Tight fit	12 GB	28
M2 Pro · 32 GB	Q3_K_L	Runs well	21 GB	23
M2 Max · 32 GB	Q3_K_L	Runs well	21 GB	35
M2 Max · 64 GB	UD-Q5_K_XL	Runs great	31 GB	31
M2 Max · 96 GB	Q8_0	Runs great	45 GB	30
M2 Ultra · 64 GB	UD-Q5_K_XL	Runs great	31 GB	43
M2 Ultra · 128 GB	Q8_0	Runs great	47 GB	44
M2 Ultra · 192 GB	Q8_0	Runs great	47 GB	44
M1 · 8 GB	UD-IQ1_M	Too big	16 GB
M1 · 16 GB	IQ2_XXS	Tight fit	12 GB	12
M1 Pro · 16 GB	IQ2_XXS	Tight fit	12 GB	22
M1 Pro · 32 GB	Q3_K_L	Runs well	21 GB	19
M1 Max · 32 GB	Q3_K_L	Runs well	21 GB	26
M1 Max · 64 GB	UD-Q5_K_XL	Runs great	31 GB	24
M1 Ultra · 64 GB	UD-Q5_K_XL	Runs great	31 GB	31
M1 Ultra · 128 GB	Q8_0	Runs great	47 GB	32

Writes is words a second, estimated from the Mac's memory speed and the model's size; how, and how wrong.

Packaged by

7 repositories on Hugging Face carry this model for llama.cpp. Outmake reads them all as one model.

Information

Maker: RecursiveSI
Released: October 2026
Parameters: 40B, 2.8B active
Longest chat: 256K tokens
Reads pictures: No
Licence: apache-2.0
Downloads this month: 469,316
Likes: 404
Sizes: 10, 11 to 80 GB
Kind: Maker's release
Source: RecursiveSI/flint

Badge for the model card

Shows on Flint's Hugging Face card which Macs run it well, linked to this page. It updates every morning.

Files, downloads and likes read from Hugging Face on November 5, 2026 and refreshed every morning. Memory and speed are estimates until someone running Outmake measures this model on that Mac.