Tier list
Local AI model tier list
Every model we track, ranked by quality alone. Tiers don’t depend on your hardware: they tell you how capable a model is. To see which ones your machine can run, use the Model Finder.
How the 70 decision models stack up
Top ratedQwen3 Reranker 4B90 Best small modelJina Reranker v3.586 Newest top-tierGLiDE 27B86 Best fully openQwen3 Reranker 8B90
Who makes the best ones · models rated A or A+
Step 1
Tell us your hardware first
Then you will see how many of these 70 decision models run on it, and can filter the list by how well each one runs.
A+
Excellent · scores 85–100
Best in class: the results people choose these models for.
9 models
A
Great · scores 70–84
Great: close to the best, often lighter or faster.
38 models
- Granite Guardian 4.1 8B 84
- Intern-Decision 4B 84
- Jina Reranker v3 84
- Llama 3.1 Nemotron Safety Guard 8B v3 84
- Qwen3Guard Gen 4B 84
- Skywork Reward V2 Qwen3 8B 84
- d1 3B 82
- mxbai-rerank-large v2 82
- FRIDA-Decisions 80
- GLiNER2.5 Decide 80
- gpt-oss-safeguard 120B 80
- Granite Guardian 3.3 8B 80
- LemonDecider 4B 80
- Qwen3 Reranker 0.6B 80
- Selene 1 Mini 8B 80
- WildGuard 7B 80
- zerank-1 small 80
- DynaGuard 8B 79
- BGE Reranker v2 Gemma 78
- CompassJudger-2 7B 78
- Nemotron Content Safety Reasoning 4B 78
- OpenAI Privacy Filter 78
- gpt-oss-safeguard 20B 76
- Llama Prompt Guard 2 86M 76
- mxbai-rerank-base v2 76
- NuExtract 2.0 4B 76
- GLiClass ModernBERT Large v3 74
- Prometheus 2 8x7B 74
- Qwen3Guard Gen 0.6B 74
- BGE Reranker v2 M3 72
- DeBERTa v3 Large Zero-shot v2.0 72
- DuoGuard 0.5B 72
- Skywork Reward V2 Qwen3 0.6B 72
- DynaGuard 1.7B 70
- GTE Reranker ModernBERT Base 70
- Jina Reranker v2 Base Multilingual 70
- ModernBERT Large Zero-shot v2.0 70
- NuExtract 2.0 2B 70
B
Good · scores 55–69
Good: dependable for most everyday jobs.
22 models
- GLiGuard 300M 68
- GLiNER2 Base v1 68
- Granite Embedding Reranker English R2 68
- Laya 68
- Llama Guard 3 8B 68
- Llama Prompt Guard 2 22M 68
- Prometheus 2 7B 68
- GLiClass Base v3 66
- GLIDER 66
- LettuceDetect Large 66
- Llama Guard 4 12B 66
- ProtectAI DeBERTa v3 Prompt Injection v2 66
- ShieldGemma 27B 66
- DeBERTa v3 Base Zero-shot v2.0 64
- Flow Judge v0.1 64
- ShieldGemma 9B 64
- GLiNER Multi v2.1 62
- HHEM 2.1 Open 62
- Laya Multilingual 60
- ShieldGemma 2B 58
- BART Large MNLI 56
- Llama Guard 3 1B 56
C
Fair · scores 40–54
Fair: usable, with visible limits.
1 model
Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.