Local AI models, ranked for phones

All 46 models we track, scored on the Galaxy S26 Ultra (16GB) as the reference flagship. The score blends fit headroom and decode speed — pick a model to see how it does on all 99 phones.

#		Model	Params	Quant · Size	Speed on Galaxy S26 Ultra	Tags	Verdict
1	S	Qwen3 0.6B	0.6B	Q8_0 · 0.6 GB	~63.6 tokens/s	chat	✓ Runs great
2	S	Qwen3 1.7B	1.7B	Q8_0 · 1.8 GB	~21.2 tokens/s	chat	✓ Runs great
3	S	Qwen3 4B	4B	Q4_K_M · 2.5 GB	~15.3 tokens/s	chat · coding	✓ Runs great
4	S	Llama 3.2 1B	1.2B	Q4_K_M · 0.8 GB	~47.7 tokens/s	chat	✓ Runs great
5	S	Llama 3.2 3B	3.2B	Q4_K_M · 2 GB	~19.1 tokens/s	chat	✓ Runs great
6	S	Gemma 3 1B	1B	Q4_K_M · 0.8 GB	~47.7 tokens/s	chat	✓ Runs great
7	S	Gemma 3 4B	4.3B	Q4_K_M · 2.5 GB	~15.3 tokens/s	chat · vision	✓ Runs great
8	S	Phi-4 Mini 3.8B	3.8B	Q4_K_M · 2.5 GB	~15.3 tokens/s	chat · reasoning	✓ Runs great
9	S	DeepSeek R1 Distill 1.5B	1.8B	Q4_K_M · 1.1 GB	~34.7 tokens/s	reasoning	✓ Runs great
10	S	SmolLM2 1.7B	1.7B	Q4_K_M · 1.1 GB	~34.7 tokens/s	chat	✓ Runs great
11	S	SmolLM3 3B	3.1B	Q4_K_M · 1.9 GB	~20.1 tokens/s	chat	✓ Runs great
12	S	Qwen 3.5 2B	2B	Q4_K_M · 1.3 GB	~29.4 tokens/s	chat · vision	✓ Runs great
13	S	Ternary Bonsai 8B	8B	PQ2_0 · 2.2 GB	~17.3 tokens/s	chat · reasoning	✓ Runs great
14	S	Ternary Bonsai 4B	4B	PQ2_0 · 1.1 GB	~34.7 tokens/s	chat	✓ Runs great
15	S	Ternary Bonsai 1.7B	1.7B	PQ2_0 · 0.5 GB	~76.3 tokens/s	chat	✓ Runs great
16	S	Ministral 3 3B	3B	Q4_K_M · 2.1 GB	~18.2 tokens/s	chat · vision	✓ Runs great
17	S	LFM2.5 8B-A1B	8B	Q4_K_M · 5.2 GB	~58.7 tokens/s	chat	✓ Runs great
18	S	Qwen 3.5 4B	4B	Q4_K_M · 2.7 GB	~14.1 tokens/s	chat · vision · reasoning	✓ Runs great
19	S	Nemotron 3 Nano 4B	4B	Q4_K_M · 2.8 GB	~13.6 tokens/s	chat · reasoning	✓ Runs great
20	S	Gemma 4 E2B	2B	Q4_K_M · 3.1 GB	~12.3 tokens/s	chat · vision	✓ Runs great
21	S	Bonsai 27B (1-bit)	27B	Q1_0 · 3.8 GB	~10 tokens/s	chat · coding · vision · reasoning	✓ Runs great
22	A	Mistral 7B v0.3	7.2B	Q4_K_M · 4.4 GB	~8.7 tokens/s	chat	✓ Runs great
23	A	DeepSeek R1 Distill 7B	7.6B	Q4_K_M · 4.7 GB	~8.1 tokens/s	reasoning	✓ Runs great
24	A	Llama 3.1 8B	8B	Q4_K_M · 4.9 GB	~7.8 tokens/s	chat · coding	! Runs, barely
25	A	Ministral 8B	8B	Q4_K_M · 4.9 GB	~7.8 tokens/s	chat	! Runs, barely
26	A	Qwen3 8B	8.2B	Q4_K_M · 5 GB	~7.6 tokens/s	chat · coding	! Runs, barely
27	A	Gemma 4 E4B	4B	Q4_K_M · 5 GB	~7.6 tokens/s	chat · vision	! Runs, barely
28	A	Ministral 3 8B	8B	Q4_K_M · 5.2 GB	~7.3 tokens/s	chat · vision	! Runs, barely
29	A	Qwen 3.5 9B	9B	Q4_K_M · 5.7 GB	~6.7 tokens/s	chat · vision · reasoning	! Runs, barely
30	A	Gemma 4 12B	12B	Q4_0 · 7 GB	~5.5 tokens/s	chat · vision · reasoning	! Runs, barely
31	A	Gemma 3 12B	12.2B	Q4_K_M · 7.3 GB	~5.2 tokens/s	chat · vision	! Runs, barely
32	A	Ternary Bonsai 27B	27B	PQ2_0 · 7.2 GB	~5.3 tokens/s	chat · coding · vision · reasoning	! Runs, barely
33	A	Ministral 3 14B	14B	Q4_K_M · 8.2 GB	~4.7 tokens/s	chat · vision · reasoning	! Runs, barely
34	B	Qwen3 14B	14.8B	Q4_K_M · 9 GB	~4.2 tokens/s	chat · coding · reasoning	! Runs, barely
35	B	Phi-4 14B	14.7B	Q4_K_M · 9.1 GB	~4.2 tokens/s	chat · reasoning	! Runs, barely
36	F	GPT-OSS 20B	21B	MXFP4 · 12.1 GB	—	chat · reasoning	✕ Won't fit
37	F	Gemma 4 26B-A4B	26B	Q4_0 · 14.4 GB	—	chat · vision · reasoning	✕ Won't fit
38	F	Qwen 3.5 27B	27B	Q4_K_M · 16.7 GB	—	chat · vision · reasoning · coding	✕ Won't fit
39	F	Qwen 3.6 27B	27B	Q4_K_M · 16.8 GB	—	chat · coding · vision · reasoning	✕ Won't fit
40	F	Gemma 4 31B	31B	Q4_K_M · 18.3 GB	—	chat · vision · reasoning	✕ Won't fit
41	F	Qwen3 30B A3B	30.5B	Q4_K_M · 18.6 GB	—	chat · coding	✕ Won't fit
42	F	Qwen3 32B	32.8B	Q4_K_M · 19.8 GB	—	chat · coding · reasoning	✕ Won't fit
43	F	Qwen 3.5 35B-A3B	35B	Q4_K_M · 22 GB	—	chat · vision · reasoning	✕ Won't fit
44	F	Qwen 3.6 35B-A3B	35B	Q4_K_M · 22.1 GB	—	chat · coding · vision · reasoning	✕ Won't fit
45	F	Nemotron 3 Nano 30B-A3B	30B	Q4_K_M · 24.6 GB	—	chat · reasoning	✕ Won't fit
46	F	Llama 3.3 70B	70B	Q4_K_M · 42.5 GB	—	chat · coding	✕ Won't fit

✓ = verified on a real device; other speeds are bandwidth-based estimates. On a phone with less RAM the ranking shifts toward smaller models — check your own phone instead.