aicanrun

AI Models for Nothing Phone (4a) Pro — What runs on 12GB

17 great · 32 slow · 44 won't fit
Chip
Snapdragon 7 Gen 4
Memory bandwidth
33.6 GB/s
NPU
—
RAM options
8 / 12 GB
Usable for models
~9 GB
Year
2026

Specs checked against manufacturer and public documentation on .

What runs on the Nothing Phone (4a) Pro

All 93 models at their recommended quant, on the 12GB configuration. Select any row for the full report.

ModelParamsQuantNeedsSpeedVerdict
Qwen3 0.6B0.6BQ8_01.3 GB~25.2 tokens/s✓ Runs great
Llama 3.2 1B1.2BQ4_K_M1.5 GB~18.9 tokens/s✓ Runs great
Gemma 3 1B1BQ4_K_M1.5 GB~18.9 tokens/s✓ Runs great
Ternary Bonsai 1.7B1.7BQ2_01.2 GB~30.2 tokens/s✓ Runs great
LFM2.5 8B-A1B8BQ4_K_M6.6 GB~23.3 tokens/s✓ Runs great
OvisOCR2 0.8B0.8BQ4_K_M1.3 GB~25.2 tokens/s✓ Runs great
Tini Cybersec 8B-A1B8.5BQ4_K_M6.7 GB~24.7 tokens/s✓ Runs great
Ling 3.0 Tiny7.9BQ4_K_M6.3 GB~17.4 tokens/s✓ Runs great
DeepSeek R1 Distill 1.5B1.8BQ4_K_M1.9 GB~13.7 tokens/s✓ Runs great
SmolLM2 1.7B1.7BQ4_K_M1.9 GB~13.7 tokens/s✓ Runs great
Ternary Bonsai 4B4BQ2_02 GB~13.7 tokens/s✓ Runs great
Qwen 3.5 2B2BQ4_K_M2.1 GB~11.6 tokens/s✓ Runs great
LFM2.5 2.6B2.7BQ4_K_M2.6 GB~8.9 tokens/s✓ Runs great
LFM2.5-VL 3B3.1BQ4_K_M2.6 GB~8.9 tokens/s✓ Runs great
Qwen3 1.7B1.7BQ8_02.6 GB~8.4 tokens/s✓ Runs great
SmolLM3 3B3.1BQ4_K_M2.8 GB~8 tokens/s✓ Runs great
G9v3 3B3BQ4_K_M2.8 GB~8 tokens/s✓ Runs great
Llama 3.2 3B3.2BQ4_K_M2.9 GB~7.6 tokens/s! Runs, barely
Ministral 3 3B3BQ4_K_M3 GB~7.2 tokens/s! Runs, barely
Ternary Bonsai 8B8BQ2_03.5 GB~6.9 tokens/s! Runs, barely
Granite 4.2 3B3BQ4_K_M3.1 GB~6.9 tokens/s! Runs, barely
Qwen3 4B4BQ4_K_M3.5 GB~6 tokens/s! Runs, barely
Gemma 3 4B4.3BQ4_K_M3.5 GB~6 tokens/s! Runs, barely
Phi-4 Mini 3.8B3.8BQ4_K_M3.5 GB~6 tokens/s! Runs, barely
Qwen 3.5 4B4BQ4_K_M3.7 GB~5.6 tokens/s! Runs, barely
Agents-A1 4B4BQ4_K_M3.7 GB~5.6 tokens/s! Runs, barely
Nanbeige 4.2 3B4.2BQ4_K_M3.7 GB~5.6 tokens/s! Runs, barely
Nemotron 3 Nano 4B4BQ4_K_M3.8 GB~5.4 tokens/s! Runs, barely
AREX Turbo 4B4.5BQ4_K_M4 GB~5.2 tokens/s! Runs, barely
Fara 1.5 4B4.5BQ4_K_M4 GB~5.2 tokens/s! Runs, barely
Gemma 4 E2B2BQ4_K_M4 GB~4.9 tokens/s! Runs, barely
Bonsai 27B (1-bit)27BQ1_04.8 GB~4 tokens/s! Runs, barely
Mistral 7B v0.37.2BQ4_K_M5.7 GB~3.4 tokens/s! Runs, barely
DeepSeek R1 Distill 7B7.6BQ4_K_M6.1 GB~3.2 tokens/s! Runs, barely
Qwen3 8B8.2BQ4_K_M6.4 GB~3 tokens/s! Runs, barely
Llama 3.1 8B8BQ4_K_M6.3 GB~3.1 tokens/s! Runs, barely
Ministral 8B8BQ4_K_M6.3 GB~3.1 tokens/s! Runs, barely
Gemma 4 E4B4BQ4_K_M6.1 GB~3 tokens/s! Runs, barely
Ministral 3 8B8BQ4_K_M6.6 GB~2.9 tokens/s! Runs, barely
Granite 4.2 8B8BQ4_K_M6.7 GB~2.9 tokens/s! Runs, barely
Ornith 1.0 9B9BQ4_K_M7.1 GB~2.7 tokens/s! Runs, barely
Qwen 3.5 9B9BQ4_K_M7.2 GB~2.7 tokens/s! Runs, barely
Qwythos 9B v29BQ4_K_M7.2 GB~2.7 tokens/s! Runs, barely
UI-Mate 9B9BQ4_K_M7.4 GB~2.6 tokens/s! Runs, barely
GRM 3.2 Cliff 9B9.4BQ4_K_M7.5 GB~2.6 tokens/s! Runs, barely
grug 9B (ProCreations)9.4BQ4_K_M7.5 GB~2.6 tokens/s! Runs, barely
Fara 1.5 9B9.4BQ4_K_M7.5 GB~2.6 tokens/s! Runs, barely
Ternary Bonsai 27B27BQ2_08.4 GB~2.1 tokens/s! Runs, barely
Gemma 4 12B12BQ4_08.8 GB~2.2 tokens/s! Runs, barely
Gemma 3 12B12.2BQ4_K_M9.1 GB—✕ Won't fit
Grug 12B12BQ4_K_M9.5 GB—✕ Won't fit
Ministral 3 14B14BQ4_K_M10.2 GB—✕ Won't fit
Qwen3 14B14.8BQ4_K_M11.1 GB—✕ Won't fit
Phi-4 14B14.7BQ4_K_M11.2 GB—✕ Won't fit
GPT-OSS 20B21BMXFP414.8 GB—✕ Won't fit
Gemma 4 26B-A4B26BQ4_017.5 GB—✕ Won't fit
Qwen3 30B A3B30.5BQ4_K_M22.3 GB—✕ Won't fit
Qwen 3.5 27B27BQ4_K_M20 GB—✕ Won't fit
Qwen 3.5 35B-A3B35BQ4_K_M26.2 GB—✕ Won't fit
Qwen 3.6 27B27BQ4_K_M18.5 GB—✕ Won't fit
Qwen 3.6 35B-A3B35BQ4_K_M23.9 GB—✕ Won't fit
Nemotron 3 Nano 30B-A3B30BQ4_K_M28.5 GB—✕ Won't fit
Nemotron 3 Nano Omni 30B-A3B31BQ4_K_M26.3 GB—✕ Won't fit
Nemotron 3.5 Lightning 30B-A3B30BQ8_035.9 GB—✕ Won't fit
Ornith 1.0 35B-A3B35BQ4_K_M25.3 GB—✕ Won't fit
XYZ-Aquila mini 35B-A3B35BQ4_K_M25.5 GB—✕ Won't fit
Salience 1.5 Flash30BQ4_K_M22.3 GB—✕ Won't fit
Fara 1.5 27B27BQ4_K_M20.9 GB—✕ Won't fit
KAT-Coder V2.5 Dev35BQ4_K_M25.5 GB—✕ Won't fit
BigBang V1 36B-A3B36BQ4_K_M26.1 GB—✕ Won't fit
Muse Glimmer 30B29.6BK_QUANT_17GB20.3 GB—✕ Won't fit
Qwen3.8 27B27BQ4_K_M20.8 GB—✕ Won't fit
Qwen3.8 2.4T-A95B2400BIQ4_XS1377.4 GB—✕ Won't fit
UI-Mate 27B27BQ4_K_M20.9 GB—✕ Won't fit
GLM-4.5V106BQ4_K_M74.8 GB—✕ Won't fit
GLM-4.5-Air106BQ4_K_M74.8 GB—✕ Won't fit
Granite 4.2 30B30BQ4_K_M21.3 GB—✕ Won't fit
Apodex 1.1 Mini35BQ4_K_M26 GB—✕ Won't fit
Ling 3.0 Flash124BQ4_K_M82.3 GB—✕ Won't fit
Qwen3.8-Flash-Next125BIQ4_XS107.7 GB—✕ Won't fit
GLM-5.3744BQ4_K_XL543.3 GB—✕ Won't fit
GLM-5.3-Flash320BQ4_K_XL232.7 GB—✕ Won't fit
Llama 3.3 70B70BQ4_K_M50.1 GB—✕ Won't fit
Qwen3 32B32.8BQ4_K_M23.7 GB—✕ Won't fit
Ornith 1.0 397B397BQ4_K_M285.6 GB—✕ Won't fit
Gemma 4 31B31BQ4_K_M22 GB—✕ Won't fit
Hunyuan 3 (Hy3)298.8BQ4_K_M212.8 GB—✕ Won't fit
DeepSeek V4 Flash 0731284BQ4_K_XL163.6 GB—✕ Won't fit
Inkling952.4BQ8_0966.9 GB—✕ Won't fit
Inkling Small276BQ4_K_M190.5 GB—✕ Won't fit
LongCat Flash Chat561.9BIQ1_S159.6 GB—✕ Won't fit
Laguna XS 2.133BQ4_K_M24.2 GB—✕ Won't fit
Laguna S 2.1118BQ4_K_M109.7 GB—✕ Won't fit

~ = bandwidth-based estimate · ✓ = measured on real hardware

Best model by use case

Best for Chat
Qwen3 0.6B

Top everyday assistant & writing pick here — ~25.2 tokens/s at Q8_0, using 1.3 of ~9GB.

Best for Coding
Tini Cybersec 8B-A1B

Top code completion & explain-this pick here — ~24.7 tokens/s at Q4_K_M, using 6.7 of ~9GB.

Best for Reasoning
Tini Cybersec 8B-A1B

Top math & step-by-step thinking pick here — ~24.7 tokens/s at Q4_K_M, using 6.7 of ~9GB.

FAQ

What is the biggest AI model the Nothing Phone (4a) Pro can run?

Bonsai 27B (1-bit) (27B parameters) at Q1_0 — it needs 4.8GB of the ~9GB usable on the 12GB Nothing Phone (4a) Pro, at ~4 tokens/s.

How much of the Nothing Phone (4a) Pro's 12GB RAM can AI models actually use?

About 9GB. Android keeps roughly 2–4GB for the system and resident apps, so of the 12GB about 9GB is actually available to a model.

Can the Nothing Phone (4a) Pro run Llama 3.1 8B?

Yes — at Q4_K_M it needs 6.3GB of the ~9GB usable and runs at ~3.1 tokens/s.

How fast is local AI on the Nothing Phone (4a) Pro?

The Snapdragon 7 Gen 4 has 33.6GB/s of memory bandwidth, which is what decode speed scales with. Small models like Ternary Bonsai 1.7B reach ~30.2 tokens/s; larger 7–14B models land in the single digits. Anything above ~8 tokens/s feels smooth for chat.

Which quantization should I use on the Nothing Phone (4a) Pro?

Q4_K_M is the size/quality sweet spot for most models. For example, Qwen3 0.6B at Q8_0 takes 1.3GB of memory here. Only drop to Q3 or IQ4 if a model just misses fitting; Q8 rarely pays off on 12GB of RAM.

Is 12GB of RAM enough for local AI?

49 of the 93 models we track fit on the Nothing Phone (4a) Pro — 17 run great and 32 run with compromises. 44 models (mostly 12B+) don't fit at their recommended quant.

Other Nothing phones

Nothing Phone (3)Nothing Phone (2a)Nothing Phone (3a)CMF Phone 2 Pro