78Fit score
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...
Model details →Best small / on-device AI models
Small models (≤ 8B params) that run on consumer GPUs or even phones. Trade-off: lower raw quality, but private, offline, and instant.
78Fit score
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...
Model details →78Fit score
Qwen3-32B is a dense 32.8B parameter causal language model from the Qwen3 series, optimized for both complex reasoning and efficient dialogue. It supports seamless switching between a "thinking" mode for...
Model details →Top 25 models for edge deployment
| # | Model | Provider | Context | Input price / 1M | Tier |
|---|---|---|---|---|---|
| 1 | Google: Gemma 4 26B A4B | 262K Long context (128K+) | $0.08 | Budget | |
| 2 | Google: Gemma 4 26B A4B (free) | 262K Long context (128K+) | $0.00 | Budget | |
| 3 | Qwen: Qwen3 32B | Qwen | 131K Long context (128K+) | $0.08 | Budget |
| 4 | LiquidAI: LFM2-8B-A1B | Liquid | 33K Short/standard context | $0.01 | Budget |
| 5 | Baidu: ERNIE 4.5 VL 28B A3B | Baidu | 30K Short/standard context | $0.14 | Budget |
| 6 | Qwen: Qwen3 8B | Qwen | 131K Long context (128K+) | $0.12 | Budget |
| 7 | Mistral: Mistral 7B Instruct v0.1 | Mistral AI | 3K Short/standard context | $0.11 | Budget |
| 8 | OpenAI: gpt-oss-120b (batch) | OpenAI | 131K Long context (128K+) | $0.03 | Budget |
| 9 | OpenAI: gpt-oss-120b (exacto) | OpenAI | 131K Long context (128K+) | $0.04 | Budget |
| 10 | OpenAI: gpt-oss-120b (free) | OpenAI | 131K Long context (128K+) | $0.00 | Budget |
| 11 | OpenAI: gpt-oss-120b | OpenAI | 131K Long context (128K+) | $0.15 | Budget |
| 12 | Arcee AI: Trinity Mini | Arcee Ai | 131K Long context (128K+) | $0.04 | Budget |
| 13 | Arcee AI: Trinity Mini (free) | Arcee Ai | 131K Long context (128K+) | $0.00 | Budget |
| 14 | Baidu: ERNIE 4.5 21B A3B | Baidu | 120K Short/standard context | $0.07 | Budget |
| 15 | Cohere: North Mini Code (free) | Cohere | 256K Long context (128K+) | $0.00 | Budget |
| 16 | Google: Gemma 3n 2B (free) | 8K Short/standard context | $0.00 | Budget | |
| 17 | meta-llama/Llama-3.2-1B-Instruct | Meta | 60K Short/standard context | $0.03 | Budget |
| 18 | meta-llama/Llama-3.2-3B-Instruct | Meta | 131K Long context (128K+) | $0.05 | Budget |
| 19 | mudler/locate-anything.cpp-gguf | Hugging Face | Unknown Unknown context | Custom | Variable |
| 20 | IBM: Granite 4.0 Micro | Ibm Granite | 131K Long context (128K+) | $0.02 | Budget |
| 21 | inclusionAI: Ling 3.0 Flash | Inclusionai | 262K Long context (128K+) | $0.02 | Budget |
| 22 | inclusionAI: Ling 3.0 Flash Fin | Inclusionai | 262K Long context (128K+) | $0.04 | Budget |
| 23 | inclusionAI: Ling 3.0 Flash Fin (free) | Inclusionai | 262K Long context (128K+) | $0.00 | Budget |
| 24 | inclusionAI: Ling 3.0 Flash Sante | Inclusionai | 262K Long context (128K+) | $0.04 | Budget |
| 25 | inclusionAI: Ling 3.0 Flash Sante (free) | Inclusionai | 262K Long context (128K+) | $0.00 | Budget |
Frequently asked questions
Can I run these models on my laptop?
Most 1–8B models run on a recent MacBook with 16GB+ RAM via Ollama or llama.cpp. Quality is well below frontier models but private and offline.
Related use cases
AI models for codingAI reasoning modelsAI vision modelslong-context AI models (128K+)AI models for function callingAI models for structured / JSON outputsAI models for agentsfree AI modelsCheapest AI models per tokenmultilingual AI modelsembedding modelsAI image generation modelsAI voice / audio modelsAI models for mathAI models for writingopen-source AI modelsAI models for RAGAI models for summarizationAI models for translationAI models for enterprise