Best local vision LLM · 2026

The best local vision LLM, ranked for your VRAM

Updated 8 Oct 2026 1 new this month

Every model on this list reads images — screenshots, photos, scanned documents, charts — and several take video or audio too. It is ordered by vendor-reported and community results on visual question answering, document reading and screen-based agent tasks (OSWorld, BabyVision) as of the catalogue date. For each memory class below we list every model on that list that actually fits at Q4_K_M with an 8K context, so the top row is the pick and the rest are what you give up by choosing it. Numbers are computed, not benchmarked; how the shortlist works.

The totals below are the language model. In GGUF the vision encoder ships as a separate projector file (mmproj), from a few hundred MB to over 1 GB, and it is not in these numbers — nor is the context each image takes, which runs from about 256 tokens to well over 1,000 per image depending on the model and the resolution. Leave headroom.

Best vision LLM for 6 GB VRAM

5.3 GB usable · speeds for a GeForce RTX 3060 Laptop everything that fits 6 GB →

Gemma 4 E4B is the pick: Laptop vision, audio too. 3 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 12K 45 Tight fit
2 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 66K 78 Runs great
3 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 71 Runs great
4 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 159 Runs great

Best vision LLM for 8 GB VRAM

7.0 GB usable · speeds for a GeForce RTX 4060 everything that fits 8 GB →

Qwen3.5 9B is the pick: Vision on an 8–12 GB card. 5 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 31K 30 Tight fit
2 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 10K 33 Tight fit
3 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 37 Runs great
4 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 120K 63 Runs great
5 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 57 Runs great
6 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 129 Runs great

Best vision LLM for 12 GB VRAM

10.6 GB usable · speeds for a GeForce RTX 3060 12 GB everything that fits 12 GB →

Gemma 4 12B is the pick: Image and audio understanding on 12 GB. 6 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 47K 32 Runs great
2 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 146K 40 Runs great
3 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 37K 43 Runs great
4 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 48 Runs great
5 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 236K 83 Runs great
6 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 76 Runs great
7 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 171 Runs great

Best vision LLM for 16 GB VRAM

14.4 GB usable · speeds for a GeForce RTX 5080 everything that fits 16 GB →

Gemma 4 12B is the pick: Image and audio understanding on 12 GB. 6 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 107K 86 Runs great
2 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 107 Runs great
3 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 66K 116 Runs great
4 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 129 Runs great
5 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 222 Runs great
6 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 203 Runs great
7 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 455 Runs great

Best vision LLM for 24 GB VRAM

22.4 GB usable · speeds for a GeForce RTX 4090 everything that fits 24 GB →

Qwen3.8 27B is the pick: Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 12 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 27.8B Q4_K_M · 16.7 GB 98K 39 Runs great
2 Gemma 4 31B
Google · Apache 2.0
Strong multimodal grounding; the most careful image reader of the dense 30Bs. 31.3B Q4_K_M · 20.2 GB 21K 35 Tight fit
3 Qwen3.6 27B
Alibaba · Apache 2.0
Qwen3.8's vision stack, one release older. 27.8B Q4_K_M · 16.7 GB 98K 39 Runs great
4 Gemma 4 26B-A4B
Google · Apache 2.0
Vision at MoE speed in 18 GB. 26.5B MoE Q4_K_M · 16.0 GB 171K 110 Runs great
5 Qwen3.6 35B-A3B
Alibaba · Apache 2.0
Fast multimodal MoE. 35.9B MoE Q4_K_M · 20.9 GB 82K 126 Tight fit
6 Muse Glimmer 30B
Meta · Apache 2.0
Screen-reading for agent loops, from Meta. 29.8B Q4_K_M · 17.5 GB 128K 36 Runs great
7 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 235K 90 Runs great
8 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 112 Runs great
9 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 126K 122 Runs great
10 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 136 Runs great
11 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 233 Runs great
12 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 213 Runs great
13 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 478 Runs great

Best vision LLM for 32 GB VRAM

30.4 GB usable · speeds for a GeForce RTX 5090 everything that fits 32 GB →

Qwen3.8 27B is the pick: Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 12 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 27.8B Q4_K_M · 16.7 GB 226K 69 Runs great
2 Gemma 4 31B
Google · Apache 2.0
Strong multimodal grounding; the most careful image reader of the dense 30Bs. 31.3B Q4_K_M · 20.2 GB 73K 62 Runs great
3 Qwen3.6 27B
Alibaba · Apache 2.0
Qwen3.8's vision stack, one release older. 27.8B Q4_K_M · 16.7 GB 226K 69 Runs great
4 Gemma 4 26B-A4B
Google · Apache 2.0
Vision at MoE speed in 18 GB. 26.5B MoE Q4_K_M · 16.0 GB 256K 195 Runs great
5 Qwen3.6 35B-A3B
Alibaba · Apache 2.0
Fast multimodal MoE. 35.9B MoE Q4_K_M · 20.9 GB 256K 225 Runs great
6 Muse Glimmer 30B
Meta · Apache 2.0
Screen-reading for agent loops, from Meta. 29.8B Q4_K_M · 17.5 GB 128K 65 Runs great
7 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 256K 161 Runs great
8 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 200 Runs great
9 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 186K 216 Runs great
10 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 241 Runs great
11 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 414 Runs great
12 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 378 Runs great
13 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 850 Runs great

Best vision LLM for 48 GB VRAM

46.4 GB usable · speeds for an RTX 6000 Ada everything that fits 48 GB →

Qwen3.8 27B is the pick: Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 12 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 27.8B Q4_K_M · 16.7 GB 256K 37 Runs great
2 Gemma 4 31B
Google · Apache 2.0
Strong multimodal grounding; the most careful image reader of the dense 30Bs. 31.3B Q4_K_M · 20.2 GB 175K 33 Runs great
3 Qwen3.6 27B
Alibaba · Apache 2.0
Qwen3.8's vision stack, one release older. 27.8B Q4_K_M · 16.7 GB 256K 37 Runs great
4 Gemma 4 26B-A4B
Google · Apache 2.0
Vision at MoE speed in 18 GB. 26.5B MoE Q4_K_M · 16.0 GB 256K 105 Runs great
5 Qwen3.6 35B-A3B
Alibaba · Apache 2.0
Fast multimodal MoE. 35.9B MoE Q4_K_M · 20.9 GB 256K 120 Runs great
6 Muse Glimmer 30B
Meta · Apache 2.0
Screen-reading for agent loops, from Meta. 29.8B Q4_K_M · 17.5 GB 128K 35 Runs great
7 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 256K 86 Runs great
8 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 107 Runs great
9 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 256K 116 Runs great
10 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 129 Runs great
11 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 222 Runs great
12 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 203 Runs great
13 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 455 Runs great

Best vision LLM for 16 GB unified memory

10.7 GB usable · speeds for an M4 · 16 GB everything that fits 16 GB →

Gemma 4 12B is the pick: Image and audio understanding on 12 GB. 6 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 48K 9.9 Runs great
2 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 149K 12 Runs great
3 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 38K 13 Runs great
4 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 15 Runs great
5 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 239K 26 Runs great
6 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 23 Runs great
7 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 53 Runs great

Best vision LLM for 64 GB unified memory

48.0 GB usable · speeds for an M4 Max · 64 GB everything that fits 64 GB →

Qwen3.8 27B is the pick: Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 12 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 27.8B Q4_K_M · 16.7 GB 256K 20 Runs great
2 Gemma 4 31B
Google · Apache 2.0
Strong multimodal grounding; the most careful image reader of the dense 30Bs. 31.3B Q4_K_M · 20.2 GB 185K 17 Runs great
3 Qwen3.6 27B
Alibaba · Apache 2.0
Qwen3.8's vision stack, one release older. 27.8B Q4_K_M · 16.7 GB 256K 20 Runs great
4 Gemma 4 26B-A4B
Google · Apache 2.0
Vision at MoE speed in 18 GB. 26.5B MoE Q4_K_M · 16.0 GB 256K 67 Runs great
5 Qwen3.6 35B-A3B
Alibaba · Apache 2.0
Fast multimodal MoE. 35.9B MoE Q4_K_M · 20.9 GB 256K 77 Runs great
6 Muse Glimmer 30B
Meta · Apache 2.0
Screen-reading for agent loops, from Meta. 29.8B Q4_K_M · 17.5 GB 128K 18 Runs great
7 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 256K 45 Runs great
8 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 56 Runs great
9 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 256K 61 Runs great
10 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 68 Runs great
11 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 116 Runs great
12 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 106 Runs great
13 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 239 Runs great

Best vision LLM for 128 GB unified memory

96.0 GB usable · speeds for an M4 Max · 128 GB everything that fits 128 GB →

Qwen3.8 27B is the pick: Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 12 other models on the vision shortlist also fit, in order of capability:

#ModelWhyParamsQuant · totalMax contextTok/s est.
1 Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. 27.8B Q4_K_M · 16.7 GB 256K 20 Runs great
2 Gemma 4 31B
Google · Apache 2.0
Strong multimodal grounding; the most careful image reader of the dense 30Bs. 31.3B Q4_K_M · 20.2 GB 256K 17 Runs great
3 Qwen3.6 27B
Alibaba · Apache 2.0
Qwen3.8's vision stack, one release older. 27.8B Q4_K_M · 16.7 GB 256K 20 Runs great
4 Gemma 4 26B-A4B
Google · Apache 2.0
Vision at MoE speed in 18 GB. 26.5B MoE Q4_K_M · 16.0 GB 256K 67 Runs great
5 Qwen3.6 35B-A3B
Alibaba · Apache 2.0
Fast multimodal MoE. 35.9B MoE Q4_K_M · 20.9 GB 256K 77 Runs great
6 Muse Glimmer 30B
Meta · Apache 2.0
Screen-reading for agent loops, from Meta. 29.8B Q4_K_M · 17.5 GB 128K 18 Runs great
7 Gemma 4 12B
Google · Apache 2.0
Image and audio understanding on 12 GB. 12B Q4_K_M · 8.2 GB 256K 45 Runs great
8 Qwen3.5 9B
Alibaba · Apache 2.0
Vision on an 8–12 GB card. 9.65B Q4_K_M · 6.3 GB 256K 56 Runs great
9 Ministral 3 8B
Mistral AI · Apache 2.0
Pixtral-lineage vision encoder in 8B. 8.92B Q4_K_M · 6.7 GB 256K 61 Runs great
10 Gemma 4 E4B
Google · Apache 2.0
Laptop vision, audio too. 8.0B Q4_K_M · 5.2 GB 128K 68 Runs great
11 Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. 4.66B Q4_K_M · 3.5 GB 256K 116 Runs great
12 Gemma 4 E2B
Google · Apache 2.0
Phone-class vision. 5.1B Q4_K_M · 3.5 GB 128K 106 Runs great
13 Qwen3.5 2B
Alibaba · Apache 2.0
Multimodal on 4 GB. 2.27B Q4_K_M · 2.0 GB 256K 239 Runs great

The whole vision shortlist, and what each one needs

Every model on the list, best first, with the smallest discrete card and the smallest Mac that run it with 15% headroom at the recommended quantisation and 8K context. Every class has its own page, Macs from 8 to 512 GB included — the full list.

#ModelParamsWeightsNeeds (GPU)Needs (Mac)Ollama tag
1 DeepSeek V4.1 Flash 552B-A16B New
Twice V4 Flash's backbone (552B, 16B active per generated token) plus 196B of Engram lookup tables (189 GiB at FP8) that DwarfStar streams from SSD, so they are not counted here. A 512 GB Mac model, and as of October 2026 no mainline llama.cpp or Ollama build: DwarfStar on a Mac, vLLM across four GPUs. DeepSeek puts the cache at 890 bytes/token; modelled conservatively as 4 latent layers plus a 128-token window.
552B MoE 310.4 GB @ Q4_K_M more than one card 512 GB Mac —
2 Kimi K3 2.8T-A104B
The largest open-weight model ever published. 69 of 93 blocks are linear attention, so its cache is tiny; its 2.8T weights are the problem.
2780B MoE 1563.2 GB @ Q4_K_M more than one card no single Mac —
3 GLM-5.3-Flash 320B-A18B
The first natively multimodal GLM-5 and the first hybrid: 34 linear-attention blocks and 11 sparse-attention blocks with a 512-wide latent cache, so a 1M window stays affordable. Z.ai says it beats GLM-5.2 at 18B active; MIT.
321B MoE 180.5 GB @ Q4_K_M more than one card 512 GB Mac —
4 Qwen3.8-Flash-Next 180B-A6B
The open preview of the Qwen4 architecture: a 125B-A6B hybrid (Gated DeltaNet + sparse attention, KV cache on 12 of 48 blocks) plus a 51B n-gram embedding and a 4B draft head — 180B on disk, 6B active. The hosted "Qwen3.8-Flash" is this model with a 1M window.
180B MoE 101.2 GB @ Q4_K_M 141 GB card 192 GB Mac qwen3.8-flash-next:125b-a6b-nvfp4
5 Qwen3.8 27B
The current default local Qwen: dense 27B, text + image + video, 262K context. Only 16 of its 64 blocks keep a KV cache, so long context is cheap.
27.8B 15.6 GB @ Q4_K_M 24 GB card 32 GB Mac qwen3.8:27b
6 Gemma 4 31B
The dense flagship: strongest maths of the 24–32 GB class (89% AIME), clean prose, vision. Q4 is a tight 24 GB fit.
31.3B 17.6 GB @ Q4_K_M 32 GB card 32 GB Mac gemma4:31b
7 Qwen3.6 27B
The 24 GB coding pick of spring 2026 (77.2 SWE-bench Verified). Same shape as 3.8, one generation behind.
27.8B 15.6 GB @ Q4_K_M 24 GB card 32 GB Mac qwen3.6:27b
8 Gemma 4 26B-A4B
Mixture of experts with 3.8B active. Slower to think than Qwen3.6 35B-A3B, faster to answer, and it sees images.
26.5B MoE 14.9 GB @ Q4_K_M 24 GB card 32 GB Mac gemma4:26b
9 Qwen3.6 35B-A3B
Mixture of experts with ~3B active: the fastest serious model a 24 GB card runs, and the best MoE under 40B on agentic coding.
35.9B MoE 20.2 GB @ Q4_K_M 32 GB card 36 GB Mac qwen3.6:35b-a3b
10 Muse Glimmer 30B
Meta's first open weights since Llama 4: a dense 30B distilled from Muse Spark for always-on local agents. Two KV heads keep the cache small.
29.8B 16.8 GB @ Q4_K_M 24 GB card 32 GB Mac muse-glimmer:30b
11 Gemma 4 12B
The "unified" Gemma 4: text, image and audio in one 12B that fits a 12 GB card at Q4. 140+ languages.
12B 6.7 GB @ Q4_K_M 11 GB card 16 GB Mac gemma4:12b
12 Qwen3.5 9B
The default for 8–12 GB cards in 2026: beats every older 8B on every published benchmark, with vision.
9.65B 5.4 GB @ Q4_K_M 10 GB card 16 GB Mac qwen3.5:9b
13 Ministral 3 8B
Mistral's 8B with images in. Plain GQA, so budget more KV cache than Qwen3.5 9B at the same context.
8.92B 5.0 GB @ Q4_K_M 10 GB card 16 GB Mac ministral-3:8b
14 Gemma 4 E4B
The laptop Gemma. 4.5B effective, 8B on disk; a single KV head per window layer keeps its cache tiny.
8.0B 4.5 GB @ Q4_K_M 8 GB card 16 GB Mac gemma4:e4b
15 Qwen3.5 4B
The 8 GB coding agent. Q4 lands near 3.4 GB, leaving room for a real context window.
4.66B 2.6 GB @ Q4_K_M 6 GB card 8 GB Mac qwen3.5:4b
16 Gemma 4 E2B
"E2B" is 2.3B effective, but the file holds 5B because of per-layer embeddings — size it as 5B. Text, image and audio in.
5.1B 2.9 GB @ Q4_K_M 6 GB card 8 GB Mac gemma4:e2b
17 Qwen3.5 2B
Phone-class, and multimodal. Replaces Llama 3.2 3B as the "it runs on anything" answer.
2.27B 1.3 GB @ Q4_K_M 6 GB card 8 GB Mac qwen3.5:2b

In llama.cpp, images go through llama-mtmd-cli or llama-server with --mmproj; Ollama and LM Studio fetch the projector for you. On a Mac, MLX runs vision models through the separate mlx-vlm package.

Other jobs