Best open-weights AI models

Open-weight models you can self-host, fine-tune, and run on-prem — ranked by meo accuracy where available. For teams with data-residency, privacy, or cost-control requirements, open weights change the build-vs-buy calculus entirely.

Ranking 162 models across the full field · as of 2026-06-08.

#	Model	Lab	Accuracy
1	DeepSeek: DeepSeek V4 Flash	deepseek	56.2%
2	MoonshotAI: Kimi K2.6	moonshotai	54.6%
3	DeepSeek: DeepSeek V4 Pro	deepseek	52.2%
4	Xiaomi: MiMo-V2.5	xiaomi	50.4%
5	NVIDIA: Nemotron 3 Ultra	nvidia	50.3%
6	Xiaomi: MiMo-V2.5-Pro	xiaomi	47.8%
7	Z.ai: GLM 5.1	z-ai	36.0%
8	Tencent: Hy3 preview	tencent	25.0%
9	Arcee AI: Trinity Large Thinking	arcee-ai	21.1%
10	AI21: Jamba Large 1.7	ai21	—
11	AionLabs: Aion-1.0-Mini	aion-labs	—
12	AllenAI: Olmo 3 32B Think	allenai	—
13	Arcee AI: Trinity Mini	arcee-ai	—
14	Baidu: ERNIE 4.5 VL 424B A47B	baidu	—
15	ByteDance: UI-TARS 7B	bytedance	—
16	Cohere: Command A	cohere	—
17	DeepSeek: DeepSeek V3	deepseek	—
18	DeepSeek: DeepSeek V3 0324	deepseek	—
19	DeepSeek: DeepSeek V3.1	deepseek	—
20	DeepSeek: DeepSeek V3.1 Terminus	deepseek	—
21	DeepSeek: DeepSeek V3.2	deepseek	—
22	DeepSeek: DeepSeek V3.2 Exp	deepseek	—
23	DeepSeek: R1	deepseek	—
24	DeepSeek: R1 0528	deepseek	—
25	DeepSeek: R1 Distill Llama 70B	deepseek	—
26	DeepSeek: R1 Distill Qwen 32B	deepseek	—
27	EssentialAI: Rnj 1 Instruct	essentialai	—
28	Google: Gemma 2 27B	google	—
29	Google: Gemma 3 12B	google	—
30	Google: Gemma 3 27B	google	—
31	Google: Gemma 3 4B	google	—
32	Google: Gemma 3n 4B	google	—
33	Google: Gemma 4 26B A4B	google	—
34	Google: Gemma 4 26B A4B (free)	google	—
35	Google: Gemma 4 31B	google	—
36	Google: Gemma 4 31B (free)	google	—
37	IBM: Granite 4.0 Micro	ibm-granite	—
38	IBM: Granite 4.1 8B	ibm-granite	—
39	LiquidAI: LFM2-24B-A2B	liquid	—
40	LiquidAI: LFM2.5-1.2B-Instruct (free)	liquid	—
41	LiquidAI: LFM2.5-1.2B-Thinking (free)	liquid	—
42	Llama Guard 3 8B	meta-llama	—
43	Magnum v4 72B	anthracite-org	—
44	Meta: Llama 3 70B Instruct	meta-llama	—
45	Meta: Llama 3 8B Instruct	meta-llama	—
46	Meta: Llama 3.1 70B Instruct	meta-llama	—
47	Meta: Llama 3.1 8B Instruct	meta-llama	—
48	Meta: Llama 3.2 11B Vision Instruct	meta-llama	—
49	Meta: Llama 3.2 1B Instruct	meta-llama	—
50	Meta: Llama 3.2 3B Instruct	meta-llama	—
51	Meta: Llama 3.2 3B Instruct (free)	meta-llama	—
52	Meta: Llama 3.3 70B Instruct	meta-llama	—
53	Meta: Llama 3.3 70B Instruct (free)	meta-llama	—
54	Meta: Llama 4 Maverick	meta-llama	—
55	Meta: Llama 4 Scout	meta-llama	—
56	Meta: Llama Guard 4 12B	meta-llama	—
57	Microsoft: Phi 4	microsoft	—
58	Microsoft: Phi 4 Mini Instruct	microsoft	—
59	MiniMax: MiniMax M2	minimax	—
60	MiniMax: MiniMax M2.1	minimax	—
61	MiniMax: MiniMax M2.5	minimax	—
62	MiniMax: MiniMax M2.7	minimax	—
63	MiniMax: MiniMax-01	minimax	—
64	Mistral: Devstral 2 2512	mistralai	—
65	Mistral: Ministral 3 14B 2512	mistralai	—
66	Mistral: Ministral 3 3B 2512	mistralai	—
67	Mistral: Ministral 3 8B 2512	mistralai	—
68	Mistral: Mistral Nemo	mistralai	—
69	Mistral: Mistral Small 3	mistralai	—
70	Mistral: Mistral Small 3.1 24B	mistralai	—
71	Mistral: Mistral Small 3.2 24B	mistralai	—
72	Mistral: Mistral Small 4	mistralai	—
73	Mistral: Mixtral 8x22B Instruct	mistralai	—
74	Mistral: Voxtral Small 24B 2507	mistralai	—
75	MoonshotAI: Kimi K2 0711	moonshotai	—
76	MoonshotAI: Kimi K2 0905	moonshotai	—
77	MoonshotAI: Kimi K2 Thinking	moonshotai	—
78	MoonshotAI: Kimi K2.5	moonshotai	—
79	MoonshotAI: Kimi K2.6 (free)	moonshotai	—
80	MythoMax 13B	gryphe	—
81	Nex AGI: DeepSeek V3.1 Nex N1	nex-agi	—
82	Nous: Hermes 3 405B Instruct	nousresearch	—
83	Nous: Hermes 3 405B Instruct (free)	nousresearch	—
84	Nous: Hermes 3 70B Instruct	nousresearch	—
85	Nous: Hermes 4 405B	nousresearch	—
86	Nous: Hermes 4 70B	nousresearch	—
87	NVIDIA: Llama 3.3 Nemotron Super 49B V1.5	nvidia	—
88	NVIDIA: Nemotron 3 Nano 30B A3B	nvidia	—
89	NVIDIA: Nemotron 3 Nano 30B A3B (free)	nvidia	—
90	NVIDIA: Nemotron 3 Super	nvidia	—
91	NVIDIA: Nemotron 3 Super (free)	nvidia	—
92	NVIDIA: Nemotron 3 Ultra (free)	nvidia	—
93	NVIDIA: Nemotron 3.5 Content Safety (free)	nvidia	—
94	NVIDIA: Nemotron Nano 12B 2 VL (free)	nvidia	—
95	NVIDIA: Nemotron Nano 9B V2	nvidia	—
96	NVIDIA: Nemotron Nano 9B V2 (free)	nvidia	—
97	OpenAI: gpt-oss-120b	openai	—
98	OpenAI: gpt-oss-120b (free)	openai	—
99	OpenAI: gpt-oss-20b	openai	—
100	OpenAI: gpt-oss-20b (free)	openai	—
101	OpenAI: gpt-oss-safeguard-20b	openai	—
102	Poolside: Laguna XS.2 (free)	poolside	—
103	Prime Intellect: INTELLECT-3	prime-intellect	—
104	Qwen: Qwen2.5 7B Instruct	qwen	—
105	Qwen: Qwen2.5 VL 72B Instruct	qwen	—
106	Qwen: Qwen3 14B	qwen	—
107	Qwen: Qwen3 235B A22B	qwen	—
108	Qwen: Qwen3 235B A22B Instruct 2507	qwen	—
109	Qwen: Qwen3 235B A22B Thinking 2507	qwen	—
110	Qwen: Qwen3 30B A3B	qwen	—
111	Qwen: Qwen3 30B A3B Instruct 2507	qwen	—
112	Qwen: Qwen3 30B A3B Thinking 2507	qwen	—
113	Qwen: Qwen3 32B	qwen	—
114	Qwen: Qwen3 8B	qwen	—
115	Qwen: Qwen3 Coder 30B A3B Instruct	qwen	—
116	Qwen: Qwen3 Coder 480B A35B	qwen	—
117	Qwen: Qwen3 Coder 480B A35B (free)	qwen	—
118	Qwen: Qwen3 Coder Next	qwen	—
119	Qwen: Qwen3 Next 80B A3B Instruct	qwen	—
120	Qwen: Qwen3 Next 80B A3B Instruct (free)	qwen	—
121	Qwen: Qwen3 Next 80B A3B Thinking	qwen	—
122	Qwen: Qwen3 VL 235B A22B Instruct	qwen	—
123	Qwen: Qwen3 VL 235B A22B Thinking	qwen	—
124	Qwen: Qwen3 VL 30B A3B Instruct	qwen	—
125	Qwen: Qwen3 VL 30B A3B Thinking	qwen	—
126	Qwen: Qwen3 VL 32B Instruct	qwen	—
127	Qwen: Qwen3 VL 8B Instruct	qwen	—
128	Qwen: Qwen3 VL 8B Thinking	qwen	—
129	Qwen: Qwen3.5 397B A17B	qwen	—
130	Qwen: Qwen3.5-122B-A10B	qwen	—
131	Qwen: Qwen3.5-27B	qwen	—
132	Qwen: Qwen3.5-35B-A3B	qwen	—
133	Qwen: Qwen3.5-9B	qwen	—
134	Qwen: Qwen3.6 27B	qwen	—
135	Qwen: Qwen3.6 35B A3B	qwen	—
136	Qwen2.5 72B Instruct	qwen	—
137	Qwen2.5 Coder 32B Instruct	qwen	—
138	Reka Edge	rekaai	—
139	Reka Flash 3	rekaai	—
140	ReMM SLERP 13B	undi95	—
141	Sao10K: Llama 3 8B Lunaris	sao10k	—
142	Sao10K: Llama 3.1 70B Hanami x1	sao10k	—
143	Sao10K: Llama 3.1 Euryale 70B v2.2	sao10k	—
144	Sao10K: Llama 3.3 Euryale 70B	sao10k	—
145	StepFun: Step 3.5 Flash	stepfun	—
146	Tencent: Hunyuan A13B Instruct	tencent	—
147	TheDrummer: Cydonia 24B V4.1	thedrummer	—
148	TheDrummer: Rocinante 12B	thedrummer	—
149	TheDrummer: Skyfall 36B V2	thedrummer	—
150	TheDrummer: UnslopNemo 12B	thedrummer	—
151	Venice: Uncensored (free)	cognitivecomputations	—
152	WizardLM-2 8x22B	microsoft	—
153	Xiaomi: MiMo-V2-Flash	xiaomi	—
154	Z.ai: GLM 4.5	z-ai	—
155	Z.ai: GLM 4.5 Air	z-ai	—
156	Z.ai: GLM 4.5 Air (free)	z-ai	—
157	Z.ai: GLM 4.5V	z-ai	—
158	Z.ai: GLM 4.6	z-ai	—
159	Z.ai: GLM 4.6V	z-ai	—
160	Z.ai: GLM 4.7	z-ai	—
161	Z.ai: GLM 4.7 Flash	z-ai	—
162	Z.ai: GLM 5	z-ai	—

← All rankings Methodology & 𝕍 →