The independent standard for
open-weight AI

Objective composite scoring and rankings for 1000 models across benchmarks, efficiency, and community usage.

1000 Models Ranked

5 Benchmarks

5 Dimensions

Free Forever

Rank	Model	Score	Tier	Benchmarks	Efficiency	Community
🥇	prism-ml Ternary-Bonsai-27B-gguf	88.4	A	93	34	64
🥈	zai-org GLM-5.2	87.3	A	91	27	72
🥉	prism-ml Bonsai-27B-gguf	86.8	A	91	34	65
#4	prism-ml Ternary-Bonsai-8B-gguf 3-13B	86.8	A	91	53	54
#5	deepseek-ai DeepSeek-V4-Pro	86.2	A	90	27	72
#6	prism-ml Bonsai-8B-gguf 3-13B	85.0	A	88	52	57
#7	Nanbeige Nanbeige4.2-3B 3-13B	84.8	A	86	74	57
#8	deepseek-ai DeepSeek-V4-Flash	84.3	A	87	27	68
#9	zai-org GLM-5	83.5	A	87	27	62
#10	zai-org GLM-5.1	83.4	A	87	27	60
#11	nvidia NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4	83.3	A	87	27	58
#12	nvidia NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16	83.1	A	87	27	59
#13	internlm internlm2_5-7b-chat 3-13B	83.1	A	86	53	52
#14	upstage Solar-Open2-250B	83.0	A	87	28	55
#15	tencent Hy3-preview	82.7	A	87	27	53
#16	zai-org GLM-4.7	82.4	A	86	27	60
#17	deepseek-ai DeepSeek-V3.2	82.4	A	85	27	67
#18	stepfun-ai Step-3.5-Flash	82.2	A	85	28	60
#19	zai-org GLM-5-FP8	82.0	A	85	27	59
#20	deepseek-ai DeepSeek-V3.1	81.8	A	85	27	61
#21	zai-org GLM-4.5	81.7	A	85	27	61
#22	moonshotai Kimi-K2-Thinking	81.4	A	84	27	59
#23	MiniMaxAI MiniMax-M2.5	80.8	A	83	28	66
#24	nvidia NVIDIA-Nemotron-3-Super-120B-A12B-BF16	80.5	A	83	28	62
#25	XiaomiMiMo MiMo-V2-Flash	80.5	A	83	27	58
#26	zai-org GLM-5.1-FP8	80.2	A	83	27	56
#27	deepseek-ai DeepSeek-V3-0324	80.2	A	81	27	70
#28	deepseek-ai DeepSeek-R1-0528	80.1	A	82	27	64
#29	Qwen Qwen3-235B-A22B-Instruct-2507	80.1	A	83	28	56
#30	MiniMaxAI MiniMax-M2	80.0	A	82	28	63
#31	ByteDance-Seed Seed-OSS-36B-Instruct	79.8	B	83	32	54
#32	meituan-longcat LongCat-Flash-Chat	79.7	B	83	27	54
#33	moonshotai Kimi-K2-Instruct	79.4	B	81	27	64
#34	deepseek-ai DeepSeek-R1	78.3	B	78	27	75
#35	nvidia NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4	77.7	B	79	36	59
#36	WeiboAI VibeThinker-3B 3-13B	77.5	B	75	81	59
#37	nvidia NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16	76.9	B	79	32	52
#38	Qwen Qwen3-30B-A3B-Thinking-2507	76.6	B	78	32	54
#39	openai gpt-oss-120b	76.5	B	76	28	72
#40	nvidia NVIDIA-Nemotron-3-Super-120B-A12B-FP8	76.0	B	77	28	56
#41	zai-org GLM-4.7-Flash	76.0	B	75	32	68
#42	zai-org GLM-4.5-Air	75.9	B	77	28	59
#43	nvidia NVIDIA-Nemotron-3-Nano-30B-A3B-FP8	75.3	B	75	32	60
#44	nvidia Nemotron-Cascade-2-30B-A3B	75.3	B	76	32	55
#45	XiaomiMiMo MiMo-V2.5-Pro	73.8	B	74	27	56
#46	meta-llama Llama-3.1-405B	73.7	B	73	27	60
#47	Qwen Qwen3-Next-80B-A3B-Instruct	73.6	B	73	29	62
#48	Qwen Qwen3-Next-80B-A3B-Thinking	73.2	B	73	29	54
#49	microsoft phi-4	72.9	B	70	37	68
#50	Qwen Qwen3-30B-A3B-Instruct-2507	72.3	B	70	32	65
#51	Qwen QwQ-32B	71.1	B	69	31	62
#52	prism-ml Bonsai-1.7B-gguf <3B Edge	71.0	B	66	95	49
#53	Qwen Qwen3-4B-Thinking-2507 3-13B	70.9	B	66	66	61
#54	Qwen Qwen3-4B-Instruct-2507 3-13B	70.8	B	66	65	65
#55	Qwen Qwen3-235B-A22B	70.2	B	68	22	64
#56	Qwen Qwen2.5-32B	70.1	B	69	31	51
#57	deepseek-ai DeepSeek-V3	70.0	B	67	22	71
#58	openai gpt-oss-20b	69.8	C	66	33	71
#59	meta-llama Llama-3.3-70B-Instruct	69.0	C	66	24	66
#60	tencent Hunyuan-A13B-Instruct	68.8	C	67	24	56
#61	nvidia NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16	67.3	C	65	31	52
#62	microsoft Phi-3.5-mini-instruct 3-13B	67.2	C	61	64	66
#63	Qwen Qwen2-72B-Instruct	66.8	C	64	24	55
#64	Qwen Qwen2.5-14B	66.6	C	64	36	52
#65	meta-llama Llama-3.1-70B-Instruct	66.6	C	63	24	65
#66	microsoft Phi-3-mini-4k-instruct 3-13B	66.0	C	59	63	65
#67	Qwen Qwen3-30B-A3B-Base	64.8	C	62	31	50
#68	XiaomiMiMo MiMo-7B-RL 3-13B	63.9	C	59	44	57
#69	ibm-granite granite-4.1-8b 3-13B	63.5	C	58	41	60
#70	LGAI-EXAONE EXAONE-3.5-32B-Instruct	62.7	C	59	31	50
#71	internlm internlm3-8b-instruct 3-13B	62.7	C	58	41	53
#72	meta-llama Meta-Llama-3-70B-Instruct	61.6	C	56	23	61
#73	microsoft Phi-4-mini-instruct 3-13B	61.2	C	53	59	63
#74	Qwen Qwen2-7B 3-13B	60.2	C	54	43	54
#75	meta-llama Meta-Llama-3-70B	59.1	D	53	23	60
#76	ibm-granite granite-4.1-3b 3-13B	58.3	D	50	62	52
#77	meta-llama Llama-3.1-70B	58.2	D	52	23	53
#78	meta-llama Llama-3.1-8B-Instruct 3-13B	58.2	D	48	40	73
#79	LiquidAI LFM2.5-8B-A1B 3-13B	57.9	D	50	40	59
#80	Qwen Qwen2-7B-Instruct 3-13B	56.3	D	47	41	62
#81	LGAI-EXAONE EXAONE-3.5-7.8B-Instruct 3-13B	55.0	D	46	40	56
#82	microsoft Phi-3-mini-128k-instruct 3-13B	54.7	D	44	54	64
#83	Qwen Qwen2.5-7B 3-13B	54.5	D	45	40	60
#84	Qwen Qwen2.5-3B 3-13B	54.2	D	44	60	56
#85	google gemma-2-9b 3-13B	54.1	D	45	37	57
#86	LiquidAI LFM2.5-1.2B-Instruct <3B Edge	53.0	D	39	91	62
#87	meta-llama Meta-Llama-3-8B-Instruct 3-13B	52.7	D	41	38	71
#88	deepseek-ai DeepSeek-Coder-V2-Lite-Instruct Code	52.0	D	42	32	62
#89	XiaomiMiMo MiMo-7B-Base 3-13B	51.6	D	42	39	53
#90	ibm-granite granite-3.1-8b-instruct 3-13B	50.8	D	41	38	52
#91	LGAI-EXAONE EXAONE-3.5-2.4B-Instruct <3B Edge	50.8	D	39	65	51
#92	meta-llama Llama-3.1-8B 3-13B	49.3	D	37	37	69
#93	meta-llama Meta-Llama-3-8B 3-13B	48.8	D	35	36	73
#94	GSAI-ML LLaDA-8B-Instruct 3-13B	48.6	D	37	37	58
#95	google gemma-2-27b-it	47.2	D	36	30	57
#96	Qwen Qwen2.5-1.5B <3B Edge	47.2	D	32	75	60
#97	google gemma-7b 3-13B	46.5	D	34	35	61
#98	HuggingFaceH4 zephyr-7b-beta 3-13B	46.1	D	33	37	62
#99	mistralai Mistral-7B-Instruct-v0.2 3-13B	45.4	D	31	36	70
#100	google gemma-2-9b-it 3-13B	45.2	D	32	34	64

Methodology

A transparent, reproducible scoring system that looks beyond simple benchmarks to capture the full picture of a model's utility.

🧠

Benchmarks

70% WEIGHT

Aggregated scores from MMLU-Pro, GPQA, HLE, GSM8K, and HumanEval. Adjusted for model contamination.

⚡

Efficiency

5% WEIGHT

Throughput (tokens/sec), memory footprint, and param-to-performance ratio on standard hardware.

🔥

Community

10% WEIGHT

Downloads, likes, and trending rank across the HuggingFace ecosystem.

🕐

Recency

15% WEIGHT

180-day half-life decay since last update. Rewards actively maintained models.

✅

Reproducibility

0% WEIGHT

Open weights, reproducible evaluation code, and clear licensing (MIT/Apache preferred). Reserved for future weighting.

Why trust ModelRank?

Open Source

MIT licensed and fully auditable methodology.
No Conflicts of Interest

We don't train or host models. Independent evaluation only.
Reproducible

All evaluation data is sourced from public HuggingFace APIs.
Updated Daily

Fully automated via GitHub Actions cron jobs.

View methodology → View source code → API docs →

Embed Badges

Showcase your model's rank anywhere with markdown.

Score Badge

![ModelRank](https://rankmodel.github.io/badges/ORG/MODEL/score.svg)

Shields.io (Works anywhere)

![ModelRank](https://img.shields.io/endpoint?url=https://rankmodel.github.io/badges/ORG/MODEL/shields.json)

Copied to clipboard!

Model

Methodology

Benchmarks

Efficiency

Community

Recency

Reproducibility

Why trust ModelRank?

Open Source

No Conflicts of Interest

Reproducible

Updated Daily

Embed Badges

Model Comparison