🛠️ RunLocalAI — local AI leaderboard & catalog

Reproducible benchmark scores and the full open-weight model catalog for running AI on your own hardware. Every benchmark is measured first-party with a public run log and a one-line reproduction command — no vibes, no leaderboard laundering.

Source of truth: runlocalai.co · Data license: CC-BY-4.0 · Click any model name for the full operator-grade page.

Ranked, reproducible quality scores on real consumer GPUs. Pick a benchmark to see the head-to-head ranking.

Benchmark


🥇	TurkishMMLU (Generative)	Turkish Llama 8B Instruct v0.1	11.8	Q4_K_M	ollama-0.24.0	81.1	⭐ First-party	run log	2026-05-28


🥇	HumanEval+ (EvalPlus)	Qwen 2.5 Coder 7B Instruct	7	Q4_K_M	ollama-0.24	81.1	⭐ First-party	run log	2026-05-28
🥈	HumanEval+ (EvalPlus)	Phi-4 14B	14	Q4_K_M	ollama-0.24	78.7	⭐ First-party	run log	2026-05-28
🥉	HumanEval+ (EvalPlus)	Trendyol LLM Asure 12B	11.8	Q4_K_M	ollama-0.24.0	69.5	⭐ First-party	run log	2026-05-27
4	HumanEval+ (EvalPlus)	Llama 3.1 8B Instruct	8	Q4_K_M	ollama-0.24	56.1	⭐ First-party	run log	2026-05-28
5	HumanEval+ (EvalPlus)	Qwen 3 8B	8	Q4_K_M	ollama-0.24	2.4	⭐ First-party	run log	2026-05-29
🥇	MBPP+ (EvalPlus)	Trendyol LLM Asure 12B	11.8	Q4_K_M	ollama-0.24.0	71.7	⭐ First-party	run log	2026-05-27
🥈	MBPP+ (EvalPlus)	Qwen 2.5 Coder 7B Instruct	7	Q4_K_M	ollama-0.24	66.9	⭐ First-party	run log	2026-05-29
🥉	MBPP+ (EvalPlus)	Phi-4 14B	14	Q4_K_M	ollama-0.24	60.3	⭐ First-party	run log	2026-05-29
4	MBPP+ (EvalPlus)	Llama 3.1 8B Instruct	8	Q4_K_M	ollama-0.24	39.2	⭐ First-party	run log	2026-05-29
🥇	TurkishMMLU (Generative)	Trendyol LLM Asure 12B	11.8	Q4_K_M	ollama-0.24.0	58.9	⭐ First-party	run log	2026-05-27
🥈	TurkishMMLU (Generative)	Llama 3.2 3B Instruct	3	Q4_K_M	ollama-0.24	11.4	⭐ First-party	run log	2026-05-28
🥉	TurkishMMLU (Generative)	Turkish Llama 8B Instruct v0.1	8	Q4_K_M	ollama-0.24	11	⭐ First-party	run log	2026-05-26
🥉	TurkishMMLU (Generative)	Turkish Llama 8B Instruct v0.1	8	Q4_K_M	ollama-0.24	11	⭐ First-party	run log	2026-05-28

What these scores mean

HumanEval+ (EvalPlus) — pass@1 /100 · coding · Liu et al., 2023 (NeurIPS). Extension of Chen et al. (2021) HumanEval. dataset
TurkishMMLU (Generative) — accuracy /100 · turkish-language, knowledge, multilingual · Yuksel et al., 2024 dataset
MBPP+ (EvalPlus) — pass@1 /100 · coding · Liu et al., 2023 (NeurIPS). Extension of Austin et al. (2021) MBPP. dataset

Every run is measured first-party on real consumer hardware and carries a public run log + a one-line reproduction command. Methodology: runlocalai.co/benchmarks/methodology.


Japanese StableLM Instruct Gamma 7B	📐 Embedding	1600B	1024K	flux-1-dev-non-commercia	⚠️ Restricted	ColPali team (Illuin Technology)	hf.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2	8.7	99


all-MiniLM-L6-v2	📐 Embedding	22M	256	apache-2.0	✅ Yes	sentence-transformers	hf.co/sentence-transformers/all-MiniLM-L6-v2	0	99
FLUX.1 [dev]	🎨 Image-gen	12B	—	flux-1-dev-non-commercia	⚠️ Restricted	Black Forest Labs	hf.co/black-forest-labs/FLUX.1-dev	0	98
DeepSeek V4 Pro (1.6T MoE)	💬 Text	1600B	1024K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-V4-Pro	0	98
Qwen 3.5 235B-A17B (MoE)	💬 Text	397B	256K	Qwen License (commercial	✅ Yes	Alibaba	hf.co/Qwen/Qwen3.5-235B-A17B	0	97
Qwen 3 235B-A22B	💬 Text	235B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-235B-A22B	0	96
Nomic Embed Text v1.5	📐 Embedding	137M	8K	apache-2.0	✅ Yes	Nomic AI	hf.co/nomic-ai/nomic-embed-text-v1.5	0	95
Llama 3.1 8B Instruct	💬 Text	8B	128K	Llama 3.1 Community Lice	✅ Yes	Meta	hf.co/meta-llama/Meta-Llama-3.1-8B-Instruct	8.7	95
DeepSeek R1 (671B reasoning)	💬 Text	671B	128K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-R1	9	95
DeepSeek V4 Flash (284B MoE)	💬 Text	284B	1024K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-V4-Flash	0	95
Llama 4 Scout	💬 Text	109B	9765K	Llama 4 Community Licens	✅ Yes	Meta	hf.co/meta-llama/Llama-4-Scout-17B-16E-Instruct	8.4	95
Kokoro 82M	🎙️ Audio	82M	—	apache-2.0	✅ Yes	Hexgrad	hf.co/hexgrad/Kokoro-82M	0	95
BGE Large EN v1.5	📐 Embedding	335M	512	mit	✅ Yes	BAAI	hf.co/BAAI/bge-large-en-v1.5	0	95
Qwen 3 0.6B	💬 Text	600M	40K	apache-2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-0.6B	0	95
Qwen 3 30B-A3B	💬 Text	30B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-30B-A3B	0	94
Qwen 2.5 Coder 32B Instruct	💬 Text	32B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen2.5-Coder-32B-Instruct	9.2	93
Llama 3.3 70B Instruct	💬 Text	70B	128K	Llama 3.3 Community Lice	✅ Yes	Meta	hf.co/meta-llama/Llama-3.3-70B-Instruct	9.1	93
BGE Reranker v2 M3	🔍 Rerank	570M	8K	MIT	✅ Yes	BAAI	hf.co/BAAI/bge-reranker-v2-m3	0	92
Qwen 3 32B	💬 Text	32B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-32B	8.9	92
Gemma 4 31B Dense	💬 Text	31B	128K	Gemma Terms of Use	✅ Yes	Google	hf.co/google/gemma-4-31b-it	0	92
all-mpnet-base-v2	📐 Embedding	109M	384	apache-2.0	✅ Yes	sentence-transformers	hf.co/sentence-transformers/all-mpnet-base-v2	0	92
XTTS v2	🎙️ Audio	460M	—	Coqui Public Model Licen	⚠️ Restricted	Coqui	hf.co/coqui/XTTS-v2	0	92
Whisper Base	🎙️ Audio	74M	30	apache-2.0	✅ Yes	OpenAI	hf.co/openai/whisper-base	0	91
Qwen 3 8B	💬 Text	8B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-8B	8.5	91
Whisper Small	🎙️ Audio	244M	30	apache-2.0	✅ Yes	OpenAI	hf.co/openai/whisper-small	0	91
Whisper Tiny	🎙️ Audio	39M	30	apache-2.0	✅ Yes	OpenAI	hf.co/openai/whisper-tiny	0	90
DeepSeek R1 Distill Llama 70B	💬 Text	70B	128K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B	9	90
paraphrase-multilingual-MiniLM-L12-v2	📐 Embedding	118M	128	apache-2.0	✅ Yes	sentence-transformers	hf.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2	0	90
Mistral Medium 3.5 (675B MoE)	💬 Text	675B	256K	Mistral Research License	⚠️ Restricted	Mistral AI	hf.co/mistralai/Mistral-Medium-3.5	0	90
DeepSeek R1 Distill Qwen 32B	💬 Text	32B	128K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B	8.8	89
GLM-5	💬 Text	200B	195K	GLM License	✅ Yes	Zhipu AI (Z.AI)	hf.co/THUDM/GLM-5	0	89
Qwen 3 1.7B	💬 Text	1.7B	40K	apache-2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-1.7B	0	88
Llama 3.2 3B Instruct	💬 Text	3B	128K	Llama 3.2 Community Lice	✅ Yes	Meta	hf.co/meta-llama/Llama-3.2-3B-Instruct	7.4	88
mxbai-embed-large-v1	📐 Embedding	335M	512	apache-2.0	✅ Yes	Mixedbread AI	hf.co/mixedbread-ai/mxbai-embed-large-v1	0	88
Jina Embeddings v3	📐 Embedding	572M	8K	cc-by-nc-4.0	⚠️ Restricted	Jina AI	hf.co/jinaai/jina-embeddings-v3	0	88
Multilingual E5 Large Instruct	📐 Embedding	560M	514	mit	✅ Yes	Microsoft (intfloat)	hf.co/intfloat/multilingual-e5-large-instruct	0	88
Gemma 4 26B MoE	💬 Text	26B	128K	Gemma Terms of Use	✅ Yes	Google	hf.co/google/gemma-4-26b-moe-it	0	88
DeepSeek V3 (671B MoE)	💬 Text	671B	64K	DeepSeek License	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-V3	9	88
Qwen 3 14B	💬 Text	14B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-14B	8.8	88
FLUX.1 [schnell]	🎨 Image-gen	12B	—	apache-2.0	✅ Yes	Black Forest Labs	hf.co/black-forest-labs/FLUX.1-schnell	0	88
Qwen2-VL 2B Instruct	👁️ Vision	2B	32K	apache-2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen2-VL-2B-Instruct	0	87
Mistral Small 3 24B	💬 Text	24B	32K	Apache 2.0	✅ Yes	Mistral AI	hf.co/mistralai/Mistral-Small-24B-Instruct-2501	8.4	87
Gemma 3 270M	💬 Text	270M	32K	gemma	✅ Yes	Google	hf.co/google/gemma-3-270m	0	87
Nemotron 3 Nano (30B-A3B)	💬 Text	30B	976K	NVIDIA Open Model Licens	✅ Yes	NVIDIA	hf.co/nvidia/Nemotron-3-Nano	0	87
Qwen 2.5 7B Instruct	💬 Text	7B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen2.5-7B-Instruct	8.6	87
Phi-4 14B	💬 Text	14B	16K	MIT	✅ Yes	Microsoft	hf.co/microsoft/phi-4	8.6	86
DeepSeek R1 Distill Qwen 7B	💬 Text	7B	128K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B	0	86
Qwen 2.5 14B Instruct	💬 Text	14B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen2.5-14B-Instruct	8.5	86
Gemma 3 27B	💬 Text	27B	128K	Gemma Terms of Use	✅ Yes	Google	hf.co/google/gemma-3-27b-it	8.2	85
Hermes 3 Llama 3.1 8B	💬 Text	8B	128K	Llama 3.1 Community Lice	✅ Yes	NousResearch	hf.co/NousResearch/Hermes-3-Llama-3.1-8B	7.7	85
Llama 3.1 70B Instruct	💬 Text	70B	128K	Llama 3.1 Community Lice	✅ Yes	Meta	hf.co/meta-llama/Meta-Llama-3.1-70B-Instruct	8	85
Qwen 3.6 35B-A3B (MTP)	💬 Text	35B	256K	Apache-2.0	✅ Yes	Alibaba / Qwen team	hf.co/Qwen/Qwen3.6-35B-A3B-MTP	8	85
Qwen 2.5 32B Instruct	💬 Text	32B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen2.5-32B-Instruct	8.8	84
Phi-4 Reasoning 14B	💬 Text	14B	32K	MIT	✅ Yes	Microsoft	hf.co/microsoft/phi-4-reasoning	8.5	84
Kimi K2.6	💬 Text	1000B	1953K	Kimi Open Weights Licens	✅ Yes	Moonshot AI	hf.co/moonshotai/Kimi-K2.6	0	84
Mistral Nemo 12B Instruct	💬 Text	12B	128K	Apache 2.0	✅ Yes	Mistral AI / NVIDIA	hf.co/mistralai/Mistral-Nemo-Instruct-2407	7.8	84
DeepSeek R1 Distill Qwen 14B	💬 Text	14B	128K	MIT	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B	0	83
SigLIP SO400M (patch14-384)	👁️ Vision	428M	—	apache-2.0	✅ Yes	Google	hf.co/google/siglip-so400m-patch14-384	0	82
Llama 3.1 Nemotron 70B Instruct	💬 Text	70B	128K	Llama 3.1 Community Lice	✅ Yes	NVIDIA	hf.co/nvidia/Llama-3.1-Nemotron-70B-Instruct	0	82
QwQ 32B Preview	💬 Text	32B	32K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/QwQ-32B-Preview	8.7	82
Llama 3.2 11B Vision Instruct	💬 Text	11B	128K	Llama 3.2 Community Lice	✅ Yes	Meta	hf.co/meta-llama/Llama-3.2-11B-Vision-Instruct	0	81
Gemma 4 E4B (Effective 4B)	💬 Text	4B	128K	Gemma Terms of Use	✅ Yes	Google	hf.co/google/gemma-4-e4b-it	0	81
Nemotron 3 Super (120B-A12B)	💬 Text	120B	976K	NVIDIA Open Model Licens	✅ Yes	NVIDIA	hf.co/nvidia/Nemotron-3-Super	0	80
Distil-Whisper Large v3	🎙️ Audio	756M	30	mit	✅ Yes	Hugging Face / Distil-Whisper	hf.co/distil-whisper/distil-large-v3	0	80
Gemma 3 12B	💬 Text	12B	128K	Gemma Terms of Use	✅ Yes	Google	hf.co/google/gemma-3-12b-it	7.9	80
Qwen 3 4B	💬 Text	4B	128K	Apache 2.0	✅ Yes	Alibaba	hf.co/Qwen/Qwen3-4B	0	80
Snowflake Arctic Embed L v2.0	📐 Embedding	568M	8K	apache-2.0	✅ Yes	Snowflake	hf.co/Snowflake/snowflake-arctic-embed-l-v2.0	0	80
SDXL Turbo	🎨 Image-gen	2.6B	—	stabilityai-non-commerci	⚠️ Restricted	Stability AI	hf.co/stabilityai/sdxl-turbo	0	80
TinyLlama 1.1B Chat v1.0	💬 Text	1.1B	2K	apache-2.0	✅ Yes	TinyLlama	hf.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0	0	80
Jina Reranker v2 Base Multilingual	🔍 Rerank	278M	1K	cc-by-nc-4.0	⚠️ Restricted	Jina AI	hf.co/jinaai/jina-reranker-v2-base-multilingual	0	80
Qwen 2.5 72B Instruct	💬 Text	72B	128K	Qwen License	✅ Yes	Alibaba	hf.co/Qwen/Qwen2.5-72B-Instruct	9	80
SmolLM2 135M Instruct	💬 Text	135M	8K	apache-2.0	✅ Yes	Hugging Face	hf.co/HuggingFaceTB/SmolLM2-135M-Instruct	0	80
OLMo 2 32B	💬 Text	32B	32K	Apache 2.0	✅ Yes	AI2 (Allen AI)	hf.co/allenai/OLMo-2-32B-Instruct	0	79
Phi-3.5 Mini Instruct	💬 Text	3.8B	128K	MIT	✅ Yes	Microsoft	hf.co/microsoft/Phi-3.5-mini-instruct	7.2	79
Qwen 3.6 27B (MTP)	💬 Text	27B	128K	Apache-2.0	✅ Yes	Alibaba / Qwen team	hf.co/Qwen/Qwen3.6-27B	8	78
DeepSeek Coder V2 Lite (16B)	💬 Text	16B	128K	DeepSeek License	✅ Yes	DeepSeek	hf.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct	8	78
Llama 4 Maverick	💬 Text	400B	976K	Llama 4 Community Licens	✅ Yes	Meta	hf.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct	8.7	78

Catalog hubs: Small LMs · Embeddings · Audio · Image · Coding · Turkish · Benchmarks

Machine-readable: models · quality-benchmarks · OpenAPI

Data licensed CC-BY-4.0 — attribute to runlocalai.co with a link.