{"version":"2026.09.08.1","reviewed":"2026-09-08","methodology":"Conservative calculated estimates for 4-bit variants at an everyday 8K context. Results include model weights, runtime overhead, context memory, execution buffers and an operating-system reserve.","models":[{"id":"llama-3.2-3b","name":"Llama 3.2 3B Instruct","parameterLabel":"3B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":3.4,"weightsGb":2,"ollamaModel":"llama3.2:3b","lmStudioRepo":"lmstudio-community/Llama-3.2-3B-Instruct-GGUF","source":"https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_2/","strengths":"A compact general assistant for older or lower-memory computers."},{"id":"qwen3-4b","name":"Qwen3 4B","parameterLabel":"4B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":4.2,"weightsGb":2.5,"ollamaModel":"qwen3:4b","lmStudioRepo":"lmstudio-community/Qwen3-4B-GGUF","source":"https://qwenlm.github.io/blog/qwen3/","strengths":"Strong everyday chat and reasoning in a small memory footprint."},{"id":"gemma-3-4b","name":"Gemma 3 4B","parameterLabel":"4B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":4.5,"weightsGb":3.3,"ollamaModel":"gemma3:4b","lmStudioRepo":"lmstudio-community/gemma-3-4b-it-GGUF","source":"https://ai.google.dev/gemma/docs/core/model_card_3","strengths":"A capable small assistant with image understanding support."},{"id":"qwen2.5-coder-7b","name":"Qwen2.5-Coder 7B","parameterLabel":"7B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":6.3,"weightsGb":4.7,"ollamaModel":"qwen2.5-coder:7b","lmStudioRepo":"lmstudio-community/Qwen2.5-Coder-7B-Instruct-GGUF","source":"https://qwenlm.github.io/blog/qwen2.5-coder-family/","strengths":"Purpose-built for code generation, explanation and completion."},{"id":"qwen3-8b","name":"Qwen3 8B","parameterLabel":"8B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":6.8,"weightsGb":5.2,"ollamaModel":"qwen3:8b","lmStudioRepo":"lmstudio-community/Qwen3-8B-GGUF","source":"https://qwenlm.github.io/blog/qwen3/","strengths":"A balanced local model for chat, code and multi-step reasoning."},{"id":"deepseek-r1-7b","name":"DeepSeek-R1 Distill Qwen 7B","parameterLabel":"7B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":6.6,"weightsGb":4.7,"ollamaModel":"deepseek-r1:7b","lmStudioRepo":"lmstudio-community/DeepSeek-R1-Distill-Qwen-7B-GGUF","source":"https://github.com/deepseek-ai/DeepSeek-R1","strengths":"A compact reasoning model for problems that benefit from working steps."},{"id":"gemma-3-12b","name":"Gemma 3 12B","parameterLabel":"12B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":10.2,"weightsGb":8.1,"ollamaModel":"gemma3:12b","lmStudioRepo":"lmstudio-community/gemma-3-12b-it-GGUF","source":"https://ai.google.dev/gemma/docs/core/model_card_3","strengths":"Higher-quality general assistance with image understanding support."},{"id":"qwen3-14b","name":"Qwen3 14B","parameterLabel":"14B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":11.4,"weightsGb":9.3,"ollamaModel":"qwen3:14b","lmStudioRepo":"lmstudio-community/Qwen3-14B-GGUF","source":"https://qwenlm.github.io/blog/qwen3/","strengths":"A quality step up for chat, coding and reasoning when memory allows."},{"id":"deepseek-r1-14b","name":"DeepSeek-R1 Distill Qwen 14B","parameterLabel":"14B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":11.5,"weightsGb":9,"ollamaModel":"deepseek-r1:14b","lmStudioRepo":"lmstudio-community/DeepSeek-R1-Distill-Qwen-14B-GGUF","source":"https://github.com/deepseek-ai/DeepSeek-R1","strengths":"A stronger local reasoning model for maths, analysis and code."},{"id":"qwen3-30b-a3b","name":"Qwen3 30B-A3B","parameterLabel":"30B MoE / 3B active","quantisation":"Q4_K_M","estimatedPeakMemoryGb":21.5,"weightsGb":18.6,"activeWeightsGb":1.9,"ollamaModel":"qwen3:30b","lmStudioRepo":"lmstudio-community/Qwen3-30B-A3B-GGUF","source":"https://qwenlm.github.io/blog/qwen3/","strengths":"A mixture-of-experts model with strong quality and relatively fast generation."},{"id":"gemma-3-27b","name":"Gemma 3 27B","parameterLabel":"27B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":22,"weightsGb":17.4,"ollamaModel":"gemma3:27b","lmStudioRepo":"lmstudio-community/gemma-3-27b-it-GGUF","source":"https://ai.google.dev/gemma/docs/core/model_card_3","strengths":"High-quality general assistance and multimodal work on larger machines."},{"id":"qwen3-32b","name":"Qwen3 32B","parameterLabel":"32B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":24.5,"weightsGb":20.2,"ollamaModel":"qwen3:32b","lmStudioRepo":"lmstudio-community/Qwen3-32B-GGUF","source":"https://qwenlm.github.io/blog/qwen3/","strengths":"A large, capable local model for demanding chat, code and reasoning."},{"id":"deepseek-r1-32b","name":"DeepSeek-R1 Distill Qwen 32B","parameterLabel":"32B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":24.7,"weightsGb":19.9,"ollamaModel":"deepseek-r1:32b","lmStudioRepo":"lmstudio-community/DeepSeek-R1-Distill-Qwen-32B-GGUF","source":"https://github.com/deepseek-ai/DeepSeek-R1","strengths":"The strongest reasoning-focused option in this initial catalogue."},{"id":"llama-3.3-70b","name":"Llama 3.3 70B Instruct","parameterLabel":"70B","quantisation":"Q4_K_M","estimatedPeakMemoryGb":49,"weightsGb":42.5,"ollamaModel":"llama3.3:70b","lmStudioRepo":"lmstudio-community/Llama-3.3-70B-Instruct-GGUF","source":"https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_3/","strengths":"A high-quality general model for workstations with substantial memory."}]}