Best Small AI Models

Small AI models (under 10B parameters) ranked by benchmark score. Lightweight models you can run locally.

Models

Providers

Open Source

$0.24

Median $/1M in

Top 3

Gemini 2.5 Pro Preview 05-06

Full Rankings

#	Model	Avg	$/1M in	Context
1	Gemini 2.5 Pro Preview 05-06🇺🇸 Google DeepMind	76.9	$1.25	1.0M
2	o4 Mini High🇺🇸 OpenAI	72.0	$1.10	200K
3	MiniMax M3🇨🇳 minimaxOpen	70.0	$0.30	1.0M
4	MiniMax M2🇨🇳 minimaxOpen	69.5	$0.26	205K
5	phi-3-small 7.4B🇺🇸 MicrosoftOpen	67.4	N/A	0K
6	Grok 3 Mini Beta🇺🇸 xAI	64.8	$0.30	131K
7	Gemini 2.0 Flash Lite🇺🇸 Google DeepMind	64.2	$0.07	1.0M
8	MiniMax M2.7🇨🇳 minimaxOpen	63.5	$0.24	205K
9	Gemini 3.5 Flash🇺🇸 Google DeepMind	62.6	$1.50	1.0M
10	phi-3-mini 3.8B🇺🇸 MicrosoftOpen	61.0	N/A	0K
11	Qwen3 4B Thinking 2507🇨🇳 AlibabaOpen	60.6	N/A	0K
12	Gemini 3 Pro🇺🇸 Google DeepMind	60.4	N/A	0K
13	GPT-5.1-Codex-Mini🇺🇸 OpenAI	60.4	$0.25	400K
14	o3 Mini High🇺🇸 OpenAI	60.4	$1.10	200K
15	Gemini 2.5 Flash Lite🇺🇸 Google DeepMind	59.1	$0.10	1.0M
16	Mixtral 8x7B🇫🇷 Mistral AIOpen	57.8	N/A	0K
17	Mixtral 8x7B Instruct🇫🇷 Mistral AIOpen	57.8	$0.54	33K
18	Gemini 3.1 Pro Preview🇺🇸 Google DeepMind	56.6	$2.00	1.0M
19	Qwen3 8B🇨🇳 Alibaba QwenOpen	56.5	$0.12	131K
20	MiniMax M2.5🇨🇳 minimaxOpen	55.1	$0.15	205K
21	GPT-5 Mini🇺🇸 OpenAI	54.3	$0.25	400K
22	Gemini 2.0 Pro🇺🇸 Google DeepMind	53.7	N/A	0K
23	Grok-3 mini🇺🇸 xAI	53.3	N/A	0K
24	Gemini 2.5 Pro🇺🇸 Google DeepMind	52.7	$1.25	1.0M
25	Qwen2.5-Coder-3B🇨🇳 Alibaba QwenOpen	52.2	N/A	0K
26	Gemini 2.5 Pro Preview 06-05🇺🇸 Google DeepMind	50.9	$1.25	1.0M
27	o4 Mini🇺🇸 OpenAI	50.9	$1.10	200K
28	DeepSeek-Coder-V2-Lite-Base🇨🇳 DeepSeekOpen	50.8	N/A	0K
29	Qwen2 7B Instruct🇨🇳 AlibabaOpen	50.5	N/A	0K
30	Gemini 2.0 Flash🇺🇸 Google DeepMind	48.0	$0.10	1.0M
31	Gemini 2.0 Flash (Dec 2024)🇺🇸 Google DeepMind	48.0	N/A	0K
32	Gemini 3 Flash Preview🇺🇸 Google DeepMind	47.7	$0.50	1.0M
33	Gemini 1.5 Flash (May 2024)🇺🇸 Google DeepMind	47.3	N/A	0K
34	Qwen2 VL 7B Instruct🇨🇳 AlibabaOpen	47.3	N/A	0K
35	Qwen3 4B Instruct 2507🇨🇳 AlibabaOpen	47.2	N/A	0K
36	Meta Llama 3 8B Instruct🇺🇸 MetaOpen	45.2	N/A	0K
37	GPT-4.1 Mini🇺🇸 OpenAI	44.5	$0.40	1.0M
38	Qwen2.5 Coder 7B Instruct🇨🇳 AlibabaOpen	44.4	N/A	0K
39	Qwen2.5 Coder 7B Instruct🇨🇳 Alibaba QwenOpen	44.4	$0.03	33K
40	Meta Llama 3 8B🇺🇸 MetaOpen	44.2	N/A	0K
41	GPT-5 Nano🇺🇸 OpenAI	43.3	$0.05	400K
42	GPT-4o-mini (2024-07-18)🇺🇸 OpenAI	43.2	$0.15	128K
43	Mistral 7B V0.1🇫🇷 Mistral AIOpen	41.6	N/A	0K
44	Gemini 1.5 Pro (May 2024)🇺🇸 Google DeepMind	41.5	N/A	0K
45	Gemini 1.5 Pro (Feb 2024)🇺🇸 Google DeepMind	41.3	N/A	0K
46	U Baichuan 2-7B Unknown	40.3	N/A	0K
47	GPT-4o-mini🇺🇸 OpenAI	39.6	$0.15	128K
48	Qwen2.5 Coder 1.5B Instruct🇨🇳 AlibabaOpen	38.8	N/A	0K
49	Gemini 2.5 Flash🇺🇸 Google DeepMind	38.5	$0.30	1.0M
50	GPT-5.4 Mini🇺🇸 OpenAI	38.3	$0.75	400K
51	Gemini 2.0 Flash Thinking (Jan 2025)🇺🇸 Google DeepMind	37.7	N/A	0K
52	Gemma 2 2b It🇺🇸 Google DeepMindOpen	36.4	N/A	0K
53	Gemma 2 9B🇺🇸 Google DeepMindOpen	36.0	$0.03	8K
54	GPT-4.1 Nano🇺🇸 OpenAI	35.2	$0.10	1.0M
55	o3 Mini🇺🇸 OpenAI	35.2	$1.10	200K
56	Qwen2.5 7B Instruct🇨🇳 Alibaba QwenOpen	35.2	$0.04	131K
57	o1-mini🇺🇸 OpenAI	34.9	N/A	0K
58	U XGen-7B Unknown	33.9	N/A	0K
59	DeepSeek R1 Distill Llama 8B🇨🇳 DeepSeekOpen	33.6	N/A	0K
60	GPT-5.4 Nano🇺🇸 OpenAI	33.6	$0.20	400K
61	DeepSeek R1 Distill Qwen 7B🇨🇳 DeepSeekOpen	32.7	N/A	0K
62	open_llama_7b🇺🇸 MetaOpen	32.0	N/A	0K
63	U Yi 6B UnknownOpen	31.4	N/A	0K
64	Llama 3 8B Instruct🇺🇸 MetaOpen	30.8	$0.14	8K
65	Phi 4 Mini Instruct🇺🇸 MicrosoftOpen	29.4	N/A	0K
66	Phi 4 Mini Instruct🇺🇸 MicrosoftOpen	29.4	$0.08	131K
67	Gemma 2B🇺🇸 Google DeepMindOpen	29.1	N/A	0K
68	Qwen-1_8B🇨🇳 Alibaba QwenOpen	28.7	N/A	0K
69	Phi 3.5 Mini Instruct🇺🇸 MicrosoftOpen	28.2	N/A	0K
70	Phi 3 Mini 4k Instruct🇺🇸 MicrosoftOpen	27.6	N/A	0K
71	Llama 3.1 8B Instruct🇺🇸 MetaOpen	27.4	$0.02	131K
72	Llama 2 7b Chat Hf🇺🇸 MetaOpen	27.3	N/A	0K
73	Qwen2.5 3B Instruct🇨🇳 Alibaba	27.2	N/A	0K
74	Llama 3.2 3B Instruct🇺🇸 MetaOpen	24.2	$0.05	131K
75	U Baichuan1-7B Unknown	23.7	N/A	0K
76	Llama 2 7b Hf🇺🇸 MetaOpen	23.6	N/A	0K
77	CodeQwen1.5-7B🇨🇳 Alibaba QwenOpen	23.1	N/A	0K
78	Hermes 2 Pro - Llama-3 8B🇺🇸 nousresearchOpen	22.1	$0.14	8K
79	Gemini 1.0 Pro🇺🇸 Google DeepMind	21.1	N/A	0K
80	Qwen2.5 Coder 0.5B Instruct🇨🇳 AlibabaOpen	20.8	N/A	0K
81	Mistral 7B Instruct V0.2🇫🇷 Mistral AIOpen	18.5	N/A	0K
82	Qwen2.5 1.5B Instruct🇨🇳 AlibabaOpen	18.4	N/A	0K
83	U RedPajama-INCITE-7B-Base Unknown	18.0	N/A	0K
84	DeepSeek Coder 6.7B🇨🇳 DeepSeekOpen	16.7	N/A	0K
85	U Magistral Small 1.1 Unknown	16.5	N/A	0K
86	Llama 3.2 1B Instruct🇺🇸 MetaOpen	14.5	$0.03	131K
87	Qwen2 1.5B Instruct🇨🇳 AlibabaOpen	14.1	N/A	0K
88	U stablelm-tuned-alpha-7b Unknown	13.6	N/A	0K
89	Mistral 7B Instruct v0.1🇫🇷 Mistral AIOpen	12.8	$0.11	3K
90	L Vicuna 7b V1.5 LMSYSOpen	10.9	N/A	0K
91	DeepSeek R1 Distill Qwen 1.5B🇨🇳 DeepSeekOpen	10.4	N/A	0K
92	Gemma 2 2b🇺🇸 Google DeepMindOpen	10.4	N/A	0K
93	Qwen2.5 0.5B Instruct🇨🇳 AlibabaOpen	10.1	N/A	0K
94	Llama 3.2 3B Instruct (free)🇺🇸 MetaOpen	8.7	Free	131K
95	Qwen2 0.5B🇨🇳 AlibabaOpen	7.2	N/A	0K
96	Qwen2 0.5B Instruct🇨🇳 AlibabaOpen	6.6	N/A	0K
97	DeepSeek Coder 1.3B🇨🇳 DeepSeekOpen	3.2	N/A	0K
98	T TinyLlama 1.1B Chat V1.0 TinyLlamaOpen	2.8	N/A	0K

About this category

Small AI models under 10 billion parameters, ranked by benchmark score. These lightweight models can run on consumer hardware and edge devices while delivering surprisingly strong performance.

Related categories

Best Open Source AI Models

Open-source AI models ranked by benchmark score. Compare Llama, Mistral, DeepSeek, Qwen, and other open-weight models.

AI Models Under $1/1M Tokens

Cheapest AI models ranked by score. All models with input pricing under $1 per million tokens, sorted by benchmark performance.

Best AI Models for Coding

AI models ranked by coding benchmarks. Compare HumanEval+, SWE-bench Verified, Aider Polyglot, and more across all providers.

Frequently asked questions

What are the best small AI models?

Small models under 10B parameters have improved dramatically. The leaderboard above ranks them by benchmark performance across coding, reasoning, and knowledge tasks.

Can small AI models run on my laptop?

Yes. Models under 10B parameters can run on modern laptops with 16GB+ RAM using tools like Ollama, llama.cpp, or LM Studio.

How do small models compare to large ones?

Small models trade some capability for speed and efficiency. The best small models now match or exceed what large models achieved just one year ago.

Back to all models