Ran it, or it didn’t happen.

Benchmark open models on your own hardware. Runs count once their event log checks out.

Recent verified runsWhereSpeed

PassedFailedNot supported

Cloud leaderboard

Every model runs on the same pinned machine. 17 models, 70 runs.

#	Model	Pass rate				Speed	Runs
1	Qwen3.5-35B-A3BQ4_K_M	96%	9/9	8/9	9/9	201.4 tok/s	3
2	Qwen3.5-27BQ4_K_M	93%	8/9	9/9	8/9	35.5 tok/s	3
3	Qwen3-30B-A3B-Instruct-2507Q4_K_M	92%	11/12	10/12	12/12	187.6 tok/s	4
4	GLM-4.7-FlashQ4_K_M	89%	9/9	6/9	9/9	201.7 tok/s	3
5	Qwen3-Coder-30B-A3B-InstructQ4_K_M	89%	9/9	7/9	8/9	186.4 tok/s	3
6	Qwen3-8BQ4_K_M	89%	9/12	11/12	12/12	120.7 tok/s	4
7	Qwen3-14BQ4_K_M	87%	13/15	11/15	15/15	67.1 tok/s	5
8	gemma-3-27b-itQ4_K_M	86%	17/18	Not supported	14/18	36.2 tok/s	6
9	gpt-oss-20bMXFP4	85%	9/9	9/9	5/9	173.8 tok/s	3
10	gemma-3-4b-itQ4_K_M	83%	11/15	Not supported	14/15	232.7 tok/s	5
11	Qwen3.5-2BQ4_K_M	81%	9/12	9/12	11/12	298.1 tok/s	4
12	Qwen3-4BQ4_K_M	78%	6/9	6/9	9/9	247.9 tok/s	3
13	gemma-3-12b-itQ4_K_M	75%	8/12	Not supported	10/12	80.9 tok/s	4
14	Llama-3.1-8B-InstructQ4_K_M	71%	8/15	9/15	15/15	124.4 tok/s	5
15	Phi-3.5-mini-instructQ4_K_M	64%	9/18	Not supported	14/18	261.3 tok/s	6
16	MiniCPM5-2BQ4_K_M	52%	4/9	3/9	7/9	301.4 tok/s	3
17	Mistral-7B-Instruct-v0.2Q4_K_M	47%	6/18	Not supported	11/18	137.1 tok/s	6

Ranked by checks passed. A task a model can’t do is shown, not counted as a zero.

143 open-weight models in creative writing ranked by LMArena score · updated Oct 28, 2026

#	Model	Score	Votes	Params	License	Downloads	Likes
143	Qwen1.5-4B-ChatQwen	969	1.1K	4.0B	Custom	15.9K	53
142	gemma-2b-itgoogle	986	696	2.5B	Gemma	13.8K	975
141	OLMo-7B-Instructallenai	989	1K	6.9B	Apache 2.0	845	54
140	gemma-1.1-2b-itgoogle	998	1.5K	2.5B	Gemma	6.1K	175
139	gemma-7b-itgoogle	1023	1.3K	8.5B	Gemma	25.3K	1,263
138	Phi-3-mini-128k-instructmicrosoft	1024	2.9K	3.8B	MIT	236.2K	1,713
137	Qwen-14B-ChatQwen	1027	898	14.2B	—	5.2K	381
136	Llama-3.2-1B-Instructmeta-llama	1032	1.1K	1.2B	Llama 3.2	1.5M	1,958
135	Llama-2-7b-chat-hfmeta-llama	1033	2.3K	6.7B	Llama 2	126.9K	4,949
134	Qwen1.5-7B-ChatQwen	1034	701	7.7B	Tongyi Qianwen	7.1K	192
133	Phi-3-mini-4k-instructmicrosoft	1044	2K	3.8B	MIT	324.8K	1,489
132	granite-3.0-2b-instructibm-granite	1046	916	2.6B	Apache 2.0	3.2K	48
131	Llama-2-13b-chat-hfmeta-llama	1047	3.3K	13.0B	Llama 2	152.7K	1,138
130	gemma-1.1-7b-itgoogle	1059	3.4K	8.5B	Gemma	4.3K	276
129	deepseek-llm-67b-chatdeepseek-ai	1067	789	—	Custom	2.7K	222
128	Mistral-7B-Instruct-v0.2mistralai	1067	3K	7.2B	Apache 2.0	1.6M	3,279
127	granite-3.0-8b-instructibm-granite	1070	895	8.2B	Apache 2.0	13.1K	209
126	Llama-2-70b-chat-hfmeta-llama	1075	6.4K	69.0B	Llama 2	4.6K	2,211
125	Phi-3-small-8k-instructmicrosoft	1083	2.4K	7.4B	MIT	7.1K	182
124	Qwen1.5-32B-ChatQwen	1083	3K	32.5B	Tongyi Qianwen	46.7K	115

Scores: LMArena, CC BY 4.0. Measured on served versions, not local builds. A model with several served variants shows its best.