data: kvKbPerToken column (exact KV-cache cost for hybrid linear-attention models)

README.md

@@ -35,6 +35,7 @@ One row per model. Each row has the model's parameter count (when disclosed; clo
 | `quantization` | string | quantization level the sizing is based on (mostly Q4_K_M, some Q8_0 siblings) |
 | `minRamGb` | number | minimum RAM/VRAM in GB to run it |
 | `estimatedLoadGb` | number | estimated memory footprint once loaded |
+| `kvKbPerToken` | number or null | exact fp16 KV-cache cost in KB per token for hybrid linear-attention models (Qwen3.5/3.6, Qwen3-Next), computed from the published HF config (full_attention layers x kv_heads x head_dim x 2 x 2 bytes; only full-attention layers cache KV); `null` means standard GQA, where KV is estimated by size class. Example: Qwen3.6 35B-A3B is 20 KB/token, so a full 262k-token fp16 cache is ~5 GB |
 | `runsLocally` | boolean | true when a registry-verified Ollama build fits a consumer RAM tier tracked here (up to 256GB) |
 | `openWeights` | boolean | true when the weights are publicly downloadable; can be true while runsLocally is false (open-weight giants like Nemotron 3 Ultra, 550B ~190GB at 2-bit, exceed every consumer tier) |
 | `ggufDiy` | boolean | true when the weights are open and a ~Q4 GGUF (0.6 GB per billion parameters) fits a 256GB-class machine via llama.cpp but no Ollama build exists (e.g. DeepSeek V4 Flash 284B, Xiaomi MiMo-V2-Flash 309B); runnable DIY, not scored for local fit |

models.csv

@@ -1,108 +1,108 @@
-model,family,params,quantization,minRamGb,estimatedLoadGb,runsLocally,openWeights,ggufDiy,runtimes,bestFor,ollamaCommand
-Qwen2.5 1.5B Instruct,Qwen,1.5,Q4_K_M,4,1.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Translation",ollama run qwen2.5:1.5b-instruct-q4_K_M
-Qwen2.5 3B Instruct,Qwen,3,Q4_K_M,6,2.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen2.5:3b-instruct-q4_K_M
-Qwen2.5 7B Instruct,Qwen,7,Q4_K_M,10,5.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen2.5:7b-instruct-q4_K_M
-Qwen2.5 14B Instruct,Qwen,14,Q4_K_M,20,11,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run qwen2.5:14b-instruct-q4_K_M
-Llama 3.2 3B Instruct,Llama,3,Q4_K_M,6,2.5,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run llama3.2:3b-instruct-q4_K_M
-Llama 3.1 8B Instruct,Llama,8,Q4_K_M,12,6.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run llama3.1:8b-instruct-q4_K_M
-Llama 3.1 8B Instruct (Q5),Llama,8,Q5_K_M,14,8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run llama3.1:8b-instruct-q5_K_M
-Llama 3.1 70B Instruct,Llama,70,Q4_K_M,48,42,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.1:70b-instruct-q4_K_M
-Mistral 7B Instruct,Mistral,7,Q4_K_M,10,5.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run mistral:7b-instruct-q4_K_M
-Mixtral 8x7B Instruct,Mistral,46.7,Q4_K_M,36,30,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run mixtral:8x7b
-Mistral Nemo 12B,Mistral,12,Q4_K_M,18,9.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Translation",ollama run mistral-nemo:12b
-Gemma 2 2B Instruct,Gemma,2,Q4_K_M,5,1.8,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run gemma2:2b-instruct-q4_K_M
-Gemma 2 9B Instruct,Gemma,9,Q4_K_M,14,7,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run gemma2:9b-instruct-q4_K_M
-Gemma 2 27B Instruct,Gemma,27,Q4_K_M,28,21,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run gemma2:27b-instruct-q4_K_M
-Phi-3 Mini 3.8B,Phi,3.8,Q4_K_M,7,3.2,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run phi3:mini
-Phi-3 Medium 14B,Phi,14,Q4_K_M,20,11,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run phi3:medium
-Phi-4 14B,Phi,14,Q4_K_M,22,11.5,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run phi4:14b-q4_K_M
-Qwen2.5 Coder 7B,Qwen,7,Q4_K_M,10,5.5,true,true,false,ollama|llama.cpp|lm-studio,Coding,ollama run qwen2.5-coder:7b
-Qwen2.5 Coder 14B,Qwen,14,Q4_K_M,22,11,true,true,false,ollama|llama.cpp|lm-studio,Coding,ollama run qwen2.5-coder:14b
-Llama 3.2 1B Instruct,Llama,1,Q4_K_M,3,1,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run llama3.2:1b-instruct-q4_K_M
-Mistral Small 22B,Mistral,22,Q4_K_M,26,17,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run mistral-small:22b
-Kimi K2 Instruct,Kimi,1000,API,0,0,false,true,false,,"Quality, Coding",
-Claude 3.5 Sonnet,Claude,,API,0,0,false,false,false,,"Coding, Chat",
-Claude 3.7 Sonnet,Claude,,API,0,0,false,false,false,,"Coding, Quality",
-Claude 3 Opus,Claude,,API,0,0,false,false,false,,"Quality, Analysis",
-Claude 4 Opus,Claude,,API,0,0,false,false,false,,"Quality, Coding",
-Qwen2.5 0.5B Instruct,Qwen,0.5,Q4_K_M,2,0.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run qwen2.5:0.5b-instruct-q4_K_M
-Gemma 3 1B Instruct,Gemma,1,Q4_K_M,3,1,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run gemma3:1b
-Phi-4 Mini 3.8B,Phi,3.8,Q4_K_M,7,3.2,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run phi4-mini:3.8b
-SmolLM2 360M,SmolLM,0.36,Q4_K_M,1,0.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Embedded",ollama run smollm2:360m
-Llama 3.3 70B Instruct,Llama,70,Q4_K_M,48,42,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q4_K_M
-Llama 3.1 405B Instruct,Llama,405,Q4_K_M,256,243,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Reasoning, Coding",ollama run llama3.1:405b-instruct-q4_K_M
-DeepSeek-R1 671B,DeepSeek,671,Q4_K_M,400,380,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding",ollama run deepseek-r1:671b-q4_K_M
-Qwen3 8B,Qwen,8,Q4_K_M,12,6.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen3:8b-q4_K_M
-Qwen3 14B,Qwen,14,Q4_K_M,20,11,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run qwen3:14b-q4_K_M
-Qwen3 30B,Qwen,30,Q4_K_M,28,22,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run qwen3:30b
-Gemma 3 4B Instruct,Gemma,4,Q4_K_M,8,3.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run gemma3:4b
-Gemma 3 12B Instruct,Gemma,12,Q4_K_M,18,9.5,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Quality",ollama run gemma3:12b
-Gemma 3 27B Instruct,Gemma,27,Q4_K_M,28,21,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run gemma3:27b
-DeepSeek-R1 Distill Qwen 7B,DeepSeek,7,Q4_K_M,10,5.5,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding",ollama run deepseek-r1:7b
-DeepSeek-R1 Distill Qwen 14B,DeepSeek,14,Q4_K_M,22,11,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Quality",ollama run deepseek-r1:14b
-DeepSeek-R1 Distill Llama 70B,DeepSeek,70,Q4_K_M,48,42,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Quality",ollama run deepseek-r1:70b
-GPT-4o,OpenAI,,API,0,0,false,false,false,,"Quality, Coding",
-GPT-4o mini,OpenAI,,API,0,0,false,false,false,,"Chat, Speed",
-Claude 4 Sonnet,Claude,,API,0,0,false,false,false,,"Coding, Quality",
-Gemini 2.5 Pro,Google,,API,0,0,false,false,false,,"Quality, Analysis",
-Gemini 2.5 Flash,Google,,API,0,0,false,false,false,,"Speed, Chat",
-DeepSeek-V3,DeepSeek,671,API,0,0,false,true,false,,"Quality, Coding",
-GLM-5,Zhipu,744,API,0,0,false,true,false,,"Reasoning, Quality",
-GLM-4 Plus,Zhipu,,API,0,0,false,false,false,,"Quality, Analysis",
-Qwen3 235B A22B,Qwen,235,Q4_K_M,192,130,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Reasoning",ollama run qwen3:235b-a22b-q4_K_M
-Mistral Small 3.1,Mistral,24,Q4_K_M,24,15,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run mistral-small3.1:24b
-DeepSeek-V3-0324,DeepSeek,671,API,0,0,false,true,false,,"Quality, Coding",
-DeepSeek-R1,DeepSeek,671,API,0,0,false,true,false,,"Reasoning, Quality",
-Qwen3.5 0.8B Instruct,Qwen,0.8,Q4_K_M,2,0.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run qwen3.5:0.8b
-Qwen3.5 2B Instruct,Qwen,2,Q4_K_M,4,1.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Edge tasks",ollama run qwen3.5:2b
-Qwen3.5 4B Instruct,Qwen,4,Q4_K_M,8,3.5,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Agents, Multimodal",ollama run qwen3.5:4b
-Qwen3.5 9B Instruct,Qwen,9,Q4_K_M,14,7,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Reasoning",ollama run qwen3.5:9b
-Qwen3.5 35B-A3B Instruct,Qwen,35,Q4_K_M,24,20,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agent scenarios",ollama run qwen3.5:35b-a3b
-Qwen3.5 27B Instruct,Qwen,27,Q4_K_M,20,16,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Complex reasoning",ollama run qwen3.5:27b
-Qwen3.5 122B-A10B Instruct,Qwen,122,Q4_K_M,96,72,true,true,false,ollama|llama.cpp|lm-studio,"Frontier-level reasoning, Complex tasks",ollama run qwen3.5:122b-a10b
-LFM2 24B-A2B Instruct,LFM2,24,Q4_K_M,16,14,true,true,false,ollama|llama.cpp|lm-studio,"Local AI agents, privacy-first tool calling, MCP workflows",ollama run lfm2:24b-a2b
-LFM2.5 8B-A1B,LFM2,8.3,Q4_K_M,10,5.5,true,true,false,ollama|llama.cpp|lm-studio,"On-device agents, tool calling, multilingual chat",ollama run lfm2.5:8b-a1b-q4_K_M
-Granite 4.1 3B Instruct,Granite,3,Q4_K_M,6,2,true,true,false,ollama|llama.cpp|lm-studio,"Lightweight chat, classification, edge tasks",ollama run granite4.1:3b
-Granite 4.1 8B Instruct,Granite,8,Q4_K_M,12,5.5,true,true,false,ollama|llama.cpp|lm-studio,"Enterprise assistant, tool calling, instruction following",ollama run granite4.1:8b
-Qwen3.6 27B,Qwen,27,Q4_K_M,24,18,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality, Long context",ollama run qwen3.6:27b
-Qwen3.6 35B-A3B,Qwen,35,Q4_K_M,24,22,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run qwen3.6:35b-a3b
-Gemma 4 31B,Gemma,31,Q4_K_M,32,20,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Multimodal",ollama run gemma4:31b
-Gemma 4 26B-A4B,Gemma,26,Q4_K_M,24,16,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:26b
-Gemma 4 E4B,Gemma,4.5,Q4_K_M,8,4,true,true,false,ollama|llama.cpp|lm-studio,"On-device, Mobile, Chat",ollama run gemma4:e4b
-Gemma 4 E2B,Gemma,2.3,Q4_K_M,4,2.3,true,true,false,ollama|llama.cpp|lm-studio,"IoT, Mobile, Edge",ollama run gemma4:e2b
-Llama 4 Scout,Llama,109,Q4_K_M,80,67,true,true,false,ollama|llama.cpp|lm-studio,"Long context, Quality, Multimodal",ollama run llama4:scout
-Llama 4 Maverick,Llama,400,Q4_K_M,256,245,true,true,false,ollama|llama.cpp|lm-studio,"Frontier quality, Long context",ollama run llama4:maverick
-Mistral Medium 3.5,Mistral,128,API,0,0,false,true,true,,"Coding, Quality, Tool calling",
-DeepSeek V4 Flash,DeepSeek,284,API,0,0,false,true,true,,"Reasoning, Coding, Long context",
-DeepSeek V4 Pro,DeepSeek,1600,API,0,0,false,true,false,,"Frontier reasoning, Coding",
-Kimi K2.6,Kimi,1000,API,0,0,false,true,false,,"Coding, Agentic tasks",
-GLM-5.1,Zhipu,744,API,0,0,false,true,false,,"Coding, Reasoning, Long context",
-Claude Opus 4.7,Claude,,API,0,0,false,false,false,,"Coding, Quality, Reasoning",
-Claude Opus 4.8,Claude,,API,0,0,false,false,false,,"Coding, Agentic, Reasoning",
-GPT-5.5,OpenAI,,API,0,0,false,false,false,,"Quality, Reasoning, Multimodal",
-Gemini 3.1 Pro,Google,,API,0,0,false,false,false,,"Multimodal, Long context, Quality",
-Grok 4.3,xAI,,API,0,0,false,false,false,,"Reasoning, Real-time data, Coding",
-Gemma 4 12B,Gemma,12,Q4_K_M,16,8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:12b
-Claude Fable 5,Claude,,API,0,0,false,false,false,,"Coding, Agentic, Reasoning",
-GLM-5.2,Zhipu,,API,0,0,false,false,false,,"Coding, Reasoning, Agentic",
-Kimi K2.7-Code,Kimi,1000,API,0,0,false,true,false,,"Coding, Agentic tasks",
-MiniMax M3,MiniMax,428,API,0,0,false,true,false,,"Coding, Agentic, Long context",
-Qwen3.7-Plus,Qwen,,API,0,0,false,false,false,,"Multimodal, Agentic, Vision",
-NVIDIA Nemotron 3 Ultra,Nemotron,550,API,0,0,false,true,false,,"Reasoning, Agentic, Long context",
-Xiaomi MiMo-V2-Flash,MiMo,309,API,0,0,false,true,true,,"Reasoning, Coding, Agentic tasks",
-NVIDIA Nemotron Cascade 2 30B-A3B,Nemotron,30,Q6_K,32,24,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Math, Agentic tasks",ollama run nemotron-cascade-2:30b
-Poolside Laguna XS.2,Laguna,33,Q4_K_M,32,23,true,true,false,ollama|llama.cpp|lm-studio,"Agentic coding, long-horizon software engineering",ollama run laguna-xs.2:q4_K_M
-Cohere North Mini Code,North,30,Q4_K_M,24,19,true,true,false,ollama|llama.cpp|lm-studio,"Agentic coding, repo-scale software engineering",ollama run north-mini-code-1.0:q4_K_M
-GPT-OSS 120B,GPT-OSS,117,MXFP4,96,65.4,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run gpt-oss:120b
-GPT-OSS 20B,GPT-OSS,21,MXFP4,24,13.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Reasoning",ollama run gpt-oss:20b
-Qwen3-Next 80B-A3B,Qwen,80,Q4_K_M,72,50.4,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Long Context",ollama run qwen3-next:80b
-Qwen3.6 35B-A3B (Q8),Qwen,35,Q8_0,64,38.7,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run qwen3.6:35b-a3b-q8_0
-Qwen3.5 35B-A3B Instruct (Q8),Qwen,35,Q8_0,64,38.7,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agent scenarios",ollama run qwen3.5:35b-a3b-q8_0
-Qwen3.5 9B Instruct (Q8),Qwen,9,Q8_0,16,10.7,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Reasoning",ollama run qwen3.5:9b-q8_0
-Qwen3.6 27B (Q8),Qwen,27,Q8_0,48,30,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality, Long context",ollama run qwen3.6:27b-q8_0
-Gemma 4 12B (Q8),Gemma,12,Q8_0,24,12.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:12b-it-q8_0
-Gemma 4 26B-A4B (Q8),Gemma,26,Q8_0,48,28.1,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:26b-a4b-it-q8_0
-Qwen3-Next 80B-A3B (Q8),Qwen,80,Q8_0,128,84.8,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Long Context",ollama run qwen3-next:80b-a3b-instruct-q8_0
-Qwen3 14B (Q8),Qwen,14,Q8_0,24,15.9,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run qwen3:14b-q8_0
-Llama 3.3 70B Instruct (Q8),Llama,70,Q8_0,128,75,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q8_0
-Llama 3.3 70B Instruct (Q6),Llama,70,Q6_K,96,57.9,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q6_K
+model,family,params,quantization,minRamGb,estimatedLoadGb,kvKbPerToken,runsLocally,openWeights,ggufDiy,runtimes,bestFor,ollamaCommand
+Qwen2.5 1.5B Instruct,Qwen,1.5,Q4_K_M,4,1.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Translation",ollama run qwen2.5:1.5b-instruct-q4_K_M
+Qwen2.5 3B Instruct,Qwen,3,Q4_K_M,6,2.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen2.5:3b-instruct-q4_K_M
+Qwen2.5 7B Instruct,Qwen,7,Q4_K_M,10,5.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen2.5:7b-instruct-q4_K_M
+Qwen2.5 14B Instruct,Qwen,14,Q4_K_M,20,11,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run qwen2.5:14b-instruct-q4_K_M
+Llama 3.2 3B Instruct,Llama,3,Q4_K_M,6,2.5,,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run llama3.2:3b-instruct-q4_K_M
+Llama 3.1 8B Instruct,Llama,8,Q4_K_M,12,6.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run llama3.1:8b-instruct-q4_K_M
+Llama 3.1 8B Instruct (Q5),Llama,8,Q5_K_M,14,8,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run llama3.1:8b-instruct-q5_K_M
+Llama 3.1 70B Instruct,Llama,70,Q4_K_M,48,42,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.1:70b-instruct-q4_K_M
+Mistral 7B Instruct,Mistral,7,Q4_K_M,10,5.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run mistral:7b-instruct-q4_K_M
+Mixtral 8x7B Instruct,Mistral,46.7,Q4_K_M,36,30,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run mixtral:8x7b
+Mistral Nemo 12B,Mistral,12,Q4_K_M,18,9.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Translation",ollama run mistral-nemo:12b
+Gemma 2 2B Instruct,Gemma,2,Q4_K_M,5,1.8,,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run gemma2:2b-instruct-q4_K_M
+Gemma 2 9B Instruct,Gemma,9,Q4_K_M,14,7,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run gemma2:9b-instruct-q4_K_M
+Gemma 2 27B Instruct,Gemma,27,Q4_K_M,28,21,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run gemma2:27b-instruct-q4_K_M
+Phi-3 Mini 3.8B,Phi,3.8,Q4_K_M,7,3.2,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run phi3:mini
+Phi-3 Medium 14B,Phi,14,Q4_K_M,20,11,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run phi3:medium
+Phi-4 14B,Phi,14,Q4_K_M,22,11.5,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run phi4:14b-q4_K_M
+Qwen2.5 Coder 7B,Qwen,7,Q4_K_M,10,5.5,,true,true,false,ollama|llama.cpp|lm-studio,Coding,ollama run qwen2.5-coder:7b
+Qwen2.5 Coder 14B,Qwen,14,Q4_K_M,22,11,,true,true,false,ollama|llama.cpp|lm-studio,Coding,ollama run qwen2.5-coder:14b
+Llama 3.2 1B Instruct,Llama,1,Q4_K_M,3,1,,true,true,false,ollama|llama.cpp|lm-studio,Chat,ollama run llama3.2:1b-instruct-q4_K_M
+Mistral Small 22B,Mistral,22,Q4_K_M,26,17,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run mistral-small:22b
+Kimi K2 Instruct,Kimi,1000,API,0,0,,false,true,false,,"Quality, Coding",
+Claude 3.5 Sonnet,Claude,,API,0,0,,false,false,false,,"Coding, Chat",
+Claude 3.7 Sonnet,Claude,,API,0,0,,false,false,false,,"Coding, Quality",
+Claude 3 Opus,Claude,,API,0,0,,false,false,false,,"Quality, Analysis",
+Claude 4 Opus,Claude,,API,0,0,,false,false,false,,"Quality, Coding",
+Qwen2.5 0.5B Instruct,Qwen,0.5,Q4_K_M,2,0.8,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run qwen2.5:0.5b-instruct-q4_K_M
+Gemma 3 1B Instruct,Gemma,1,Q4_K_M,3,1,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run gemma3:1b
+Phi-4 Mini 3.8B,Phi,3.8,Q4_K_M,7,3.2,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Chat",ollama run phi4-mini:3.8b
+SmolLM2 360M,SmolLM,0.36,Q4_K_M,1,0.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Embedded",ollama run smollm2:360m
+Llama 3.3 70B Instruct,Llama,70,Q4_K_M,48,42,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q4_K_M
+Llama 3.1 405B Instruct,Llama,405,Q4_K_M,256,243,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Reasoning, Coding",ollama run llama3.1:405b-instruct-q4_K_M
+DeepSeek-R1 671B,DeepSeek,671,Q4_K_M,400,380,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding",ollama run deepseek-r1:671b-q4_K_M
+Qwen3 8B,Qwen,8,Q4_K_M,12,6.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run qwen3:8b-q4_K_M
+Qwen3 14B,Qwen,14,Q4_K_M,20,11,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run qwen3:14b-q4_K_M
+Qwen3 30B,Qwen,30,Q4_K_M,28,22,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run qwen3:30b
+Gemma 3 4B Instruct,Gemma,4,Q4_K_M,8,3.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run gemma3:4b
+Gemma 3 12B Instruct,Gemma,12,Q4_K_M,18,9.5,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Quality",ollama run gemma3:12b
+Gemma 3 27B Instruct,Gemma,27,Q4_K_M,28,21,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run gemma3:27b
+DeepSeek-R1 Distill Qwen 7B,DeepSeek,7,Q4_K_M,10,5.5,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding",ollama run deepseek-r1:7b
+DeepSeek-R1 Distill Qwen 14B,DeepSeek,14,Q4_K_M,22,11,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Quality",ollama run deepseek-r1:14b
+DeepSeek-R1 Distill Llama 70B,DeepSeek,70,Q4_K_M,48,42,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Quality",ollama run deepseek-r1:70b
+GPT-4o,OpenAI,,API,0,0,,false,false,false,,"Quality, Coding",
+GPT-4o mini,OpenAI,,API,0,0,,false,false,false,,"Chat, Speed",
+Claude 4 Sonnet,Claude,,API,0,0,,false,false,false,,"Coding, Quality",
+Gemini 2.5 Pro,Google,,API,0,0,,false,false,false,,"Quality, Analysis",
+Gemini 2.5 Flash,Google,,API,0,0,,false,false,false,,"Speed, Chat",
+DeepSeek-V3,DeepSeek,671,API,0,0,,false,true,false,,"Quality, Coding",
+GLM-5,Zhipu,744,API,0,0,,false,true,false,,"Reasoning, Quality",
+GLM-4 Plus,Zhipu,,API,0,0,,false,false,false,,"Quality, Analysis",
+Qwen3 235B A22B,Qwen,235,Q4_K_M,192,130,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Reasoning",ollama run qwen3:235b-a22b-q4_K_M
+Mistral Small 3.1,Mistral,24,Q4_K_M,24,15,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding",ollama run mistral-small3.1:24b
+DeepSeek-V3-0324,DeepSeek,671,API,0,0,,false,true,false,,"Quality, Coding",
+DeepSeek-R1,DeepSeek,671,API,0,0,,false,true,false,,"Reasoning, Quality",
+Qwen3.5 0.8B Instruct,Qwen,0.8,Q4_K_M,2,0.8,12,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Mobile",ollama run qwen3.5:0.8b
+Qwen3.5 2B Instruct,Qwen,2,Q4_K_M,4,1.8,12,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Edge tasks",ollama run qwen3.5:2b
+Qwen3.5 4B Instruct,Qwen,4,Q4_K_M,8,3.5,32,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Agents, Multimodal",ollama run qwen3.5:4b
+Qwen3.5 9B Instruct,Qwen,9,Q4_K_M,14,7,32,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Reasoning",ollama run qwen3.5:9b
+Qwen3.5 35B-A3B Instruct,Qwen,35,Q4_K_M,24,20,20,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agent scenarios",ollama run qwen3.5:35b-a3b
+Qwen3.5 27B Instruct,Qwen,27,Q4_K_M,20,16,64,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Complex reasoning",ollama run qwen3.5:27b
+Qwen3.5 122B-A10B Instruct,Qwen,122,Q4_K_M,96,72,24,true,true,false,ollama|llama.cpp|lm-studio,"Frontier-level reasoning, Complex tasks",ollama run qwen3.5:122b-a10b
+LFM2 24B-A2B Instruct,LFM2,24,Q4_K_M,16,14,,true,true,false,ollama|llama.cpp|lm-studio,"Local AI agents, privacy-first tool calling, MCP workflows",ollama run lfm2:24b-a2b
+LFM2.5 8B-A1B,LFM2,8.3,Q4_K_M,10,5.5,,true,true,false,ollama|llama.cpp|lm-studio,"On-device agents, tool calling, multilingual chat",ollama run lfm2.5:8b-a1b-q4_K_M
+Granite 4.1 3B Instruct,Granite,3,Q4_K_M,6,2,,true,true,false,ollama|llama.cpp|lm-studio,"Lightweight chat, classification, edge tasks",ollama run granite4.1:3b
+Granite 4.1 8B Instruct,Granite,8,Q4_K_M,12,5.5,,true,true,false,ollama|llama.cpp|lm-studio,"Enterprise assistant, tool calling, instruction following",ollama run granite4.1:8b
+Qwen3.6 27B,Qwen,27,Q4_K_M,24,18,64,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality, Long context",ollama run qwen3.6:27b
+Qwen3.6 35B-A3B,Qwen,35,Q4_K_M,24,22,20,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run qwen3.6:35b-a3b
+Gemma 4 31B,Gemma,31,Q4_K_M,32,20,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Multimodal",ollama run gemma4:31b
+Gemma 4 26B-A4B,Gemma,26,Q4_K_M,24,16,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:26b
+Gemma 4 E4B,Gemma,4.5,Q4_K_M,8,4,,true,true,false,ollama|llama.cpp|lm-studio,"On-device, Mobile, Chat",ollama run gemma4:e4b
+Gemma 4 E2B,Gemma,2.3,Q4_K_M,4,2.3,,true,true,false,ollama|llama.cpp|lm-studio,"IoT, Mobile, Edge",ollama run gemma4:e2b
+Llama 4 Scout,Llama,109,Q4_K_M,80,67,,true,true,false,ollama|llama.cpp|lm-studio,"Long context, Quality, Multimodal",ollama run llama4:scout
+Llama 4 Maverick,Llama,400,Q4_K_M,256,245,,true,true,false,ollama|llama.cpp|lm-studio,"Frontier quality, Long context",ollama run llama4:maverick
+Mistral Medium 3.5,Mistral,128,API,0,0,,false,true,true,,"Coding, Quality, Tool calling",
+DeepSeek V4 Flash,DeepSeek,284,API,0,0,,false,true,true,,"Reasoning, Coding, Long context",
+DeepSeek V4 Pro,DeepSeek,1600,API,0,0,,false,true,false,,"Frontier reasoning, Coding",
+Kimi K2.6,Kimi,1000,API,0,0,,false,true,false,,"Coding, Agentic tasks",
+GLM-5.1,Zhipu,744,API,0,0,,false,true,false,,"Coding, Reasoning, Long context",
+Claude Opus 4.7,Claude,,API,0,0,,false,false,false,,"Coding, Quality, Reasoning",
+Claude Opus 4.8,Claude,,API,0,0,,false,false,false,,"Coding, Agentic, Reasoning",
+GPT-5.5,OpenAI,,API,0,0,,false,false,false,,"Quality, Reasoning, Multimodal",
+Gemini 3.1 Pro,Google,,API,0,0,,false,false,false,,"Multimodal, Long context, Quality",
+Grok 4.3,xAI,,API,0,0,,false,false,false,,"Reasoning, Real-time data, Coding",
+Gemma 4 12B,Gemma,12,Q4_K_M,16,8,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:12b
+Claude Fable 5,Claude,,API,0,0,,false,false,false,,"Coding, Agentic, Reasoning",
+GLM-5.2,Zhipu,,API,0,0,,false,false,false,,"Coding, Reasoning, Agentic",
+Kimi K2.7-Code,Kimi,1000,API,0,0,,false,true,false,,"Coding, Agentic tasks",
+MiniMax M3,MiniMax,428,API,0,0,,false,true,false,,"Coding, Agentic, Long context",
+Qwen3.7-Plus,Qwen,,API,0,0,,false,false,false,,"Multimodal, Agentic, Vision",
+NVIDIA Nemotron 3 Ultra,Nemotron,550,API,0,0,,false,true,false,,"Reasoning, Agentic, Long context",
+Xiaomi MiMo-V2-Flash,MiMo,309,API,0,0,,false,true,true,,"Reasoning, Coding, Agentic tasks",
+NVIDIA Nemotron Cascade 2 30B-A3B,Nemotron,30,Q6_K,32,24,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Math, Agentic tasks",ollama run nemotron-cascade-2:30b
+Poolside Laguna XS.2,Laguna,33,Q4_K_M,32,23,,true,true,false,ollama|llama.cpp|lm-studio,"Agentic coding, long-horizon software engineering",ollama run laguna-xs.2:q4_K_M
+Cohere North Mini Code,North,30,Q4_K_M,24,19,,true,true,false,ollama|llama.cpp|lm-studio,"Agentic coding, repo-scale software engineering",ollama run north-mini-code-1.0:q4_K_M
+GPT-OSS 120B,GPT-OSS,117,MXFP4,96,65.4,,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run gpt-oss:120b
+GPT-OSS 20B,GPT-OSS,21,MXFP4,24,13.8,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Reasoning",ollama run gpt-oss:20b
+Qwen3-Next 80B-A3B,Qwen,80,Q4_K_M,72,50.4,24,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Long Context",ollama run qwen3-next:80b
+Qwen3.6 35B-A3B (Q8),Qwen,35,Q8_0,64,38.7,20,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agents",ollama run qwen3.6:35b-a3b-q8_0
+Qwen3.5 35B-A3B Instruct (Q8),Qwen,35,Q8_0,64,38.7,20,true,true,false,ollama|llama.cpp|lm-studio,"Reasoning, Coding, Agent scenarios",ollama run qwen3.5:35b-a3b-q8_0
+Qwen3.5 9B Instruct (Q8),Qwen,9,Q8_0,16,10.7,32,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding, Reasoning",ollama run qwen3.5:9b-q8_0
+Qwen3.6 27B (Q8),Qwen,27,Q8_0,48,30,64,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality, Long context",ollama run qwen3.6:27b-q8_0
+Gemma 4 12B (Q8),Gemma,12,Q8_0,24,12.8,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:12b-it-q8_0
+Gemma 4 26B-A4B (Q8),Gemma,26,Q8_0,48,28.1,,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Multimodal",ollama run gemma4:26b-a4b-it-q8_0
+Qwen3-Next 80B-A3B (Q8),Qwen,80,Q8_0,128,84.8,24,true,true,false,ollama|llama.cpp|lm-studio,"Chat, Coding, Long Context",ollama run qwen3-next:80b-a3b-instruct-q8_0
+Qwen3 14B (Q8),Qwen,14,Q8_0,24,15.9,,true,true,false,ollama|llama.cpp|lm-studio,"Coding, Quality",ollama run qwen3:14b-q8_0
+Llama 3.3 70B Instruct (Q8),Llama,70,Q8_0,128,75,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q8_0
+Llama 3.3 70B Instruct (Q6),Llama,70,Q6_K,96,57.9,,true,true,false,ollama|llama.cpp|lm-studio,"Quality, Coding",ollama run llama3.3:70b-instruct-q6_K

models.json

@@ -10,7 +10,7 @@
     "note": "A model fits a device when its estimatedLoadGb is within the memory budget: ~70% of unified memory up to 32GB, scaling linearly to ~85% at 128GB and above (high-RAM Macs can wire more memory to the GPU). Q4_K_M needs roughly 0.6 GB per billion parameters.",
     "fieldSemantics": "runsLocally = a registry-verified Ollama build fits at least one consumer RAM tier tracked here (up to 256GB). openWeights = the weights are publicly downloadable. Some open-weight giants (e.g. NVIDIA Nemotron 3 Ultra, 550B, ~190GB at 2-bit) are openWeights: true but runsLocally: false because no consumer tier fits them."
   },
-  "updated": "2026-07-05",
+  "updated": "2026-07-06",
   "counts": {
     "models": 107,
     "local": 75,
@@ -25,6 +25,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 4,
       "estimatedLoadGb": 1.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -43,6 +44,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 6,
       "estimatedLoadGb": 2.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -61,6 +63,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 10,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -79,6 +82,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 20,
       "estimatedLoadGb": 11,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -97,6 +101,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 6,
       "estimatedLoadGb": 2.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -115,6 +120,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 12,
       "estimatedLoadGb": 6.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -133,6 +139,7 @@
       "quantization": "Q5_K_M",
       "minRamGb": 14,
       "estimatedLoadGb": 8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -151,6 +158,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 48,
       "estimatedLoadGb": 42,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -169,6 +177,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 10,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -187,6 +196,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 36,
       "estimatedLoadGb": 30,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -205,6 +215,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 18,
       "estimatedLoadGb": 9.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -223,6 +234,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 5,
       "estimatedLoadGb": 1.8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -241,6 +253,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 14,
       "estimatedLoadGb": 7,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -259,6 +272,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 28,
       "estimatedLoadGb": 21,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -277,6 +291,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 7,
       "estimatedLoadGb": 3.2,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -295,6 +310,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 20,
       "estimatedLoadGb": 11,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -313,6 +329,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 22,
       "estimatedLoadGb": 11.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -331,6 +348,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 10,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -349,6 +367,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 22,
       "estimatedLoadGb": 11,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -367,6 +386,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 3,
       "estimatedLoadGb": 1,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -385,6 +405,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 26,
       "estimatedLoadGb": 17,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -403,6 +424,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -417,6 +439,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -431,6 +454,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -445,6 +469,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -459,6 +484,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -473,6 +499,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 2,
       "estimatedLoadGb": 0.8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -491,6 +518,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 3,
       "estimatedLoadGb": 1,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -509,6 +537,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 7,
       "estimatedLoadGb": 3.2,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -527,6 +556,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 1,
       "estimatedLoadGb": 0.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -545,6 +575,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 48,
       "estimatedLoadGb": 42,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -563,6 +594,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 256,
       "estimatedLoadGb": 243,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -581,6 +613,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 400,
       "estimatedLoadGb": 380,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -599,6 +632,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 12,
       "estimatedLoadGb": 6.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -617,6 +651,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 20,
       "estimatedLoadGb": 11,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -635,6 +670,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 28,
       "estimatedLoadGb": 22,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -653,6 +689,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 8,
       "estimatedLoadGb": 3.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -671,6 +708,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 18,
       "estimatedLoadGb": 9.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -689,6 +727,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 28,
       "estimatedLoadGb": 21,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -707,6 +746,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 10,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -725,6 +765,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 22,
       "estimatedLoadGb": 11,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -743,6 +784,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 48,
       "estimatedLoadGb": 42,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -761,6 +803,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -775,6 +818,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -789,6 +833,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -803,6 +848,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -817,6 +863,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -831,6 +878,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -845,6 +893,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -859,6 +908,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -873,6 +923,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 192,
       "estimatedLoadGb": 130,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -891,6 +942,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 15,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -909,6 +961,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -923,6 +976,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -937,6 +991,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 2,
       "estimatedLoadGb": 0.8,
+      "kvKbPerToken": 12,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -955,6 +1010,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 4,
       "estimatedLoadGb": 1.8,
+      "kvKbPerToken": 12,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -973,6 +1029,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 8,
       "estimatedLoadGb": 3.5,
+      "kvKbPerToken": 32,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -991,6 +1048,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 14,
       "estimatedLoadGb": 7,
+      "kvKbPerToken": 32,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1009,6 +1067,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 20,
+      "kvKbPerToken": 20,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1027,6 +1086,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 20,
       "estimatedLoadGb": 16,
+      "kvKbPerToken": 64,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1045,6 +1105,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 96,
       "estimatedLoadGb": 72,
+      "kvKbPerToken": 24,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1063,6 +1124,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 16,
       "estimatedLoadGb": 14,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1081,6 +1143,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 10,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1099,6 +1162,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 6,
       "estimatedLoadGb": 2,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1117,6 +1181,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 12,
       "estimatedLoadGb": 5.5,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1135,6 +1200,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 18,
+      "kvKbPerToken": 64,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1153,6 +1219,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 22,
+      "kvKbPerToken": 20,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1171,6 +1238,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 32,
       "estimatedLoadGb": 20,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1189,6 +1257,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 16,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1207,6 +1276,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 8,
       "estimatedLoadGb": 4,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1225,6 +1295,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 4,
       "estimatedLoadGb": 2.3,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1243,6 +1314,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 80,
       "estimatedLoadGb": 67,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1261,6 +1333,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 256,
       "estimatedLoadGb": 245,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1279,6 +1352,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": true,
@@ -1293,6 +1367,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": true,
@@ -1307,6 +1382,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1321,6 +1397,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1335,6 +1412,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1349,6 +1427,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1363,6 +1442,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1377,6 +1457,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1391,6 +1472,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1405,6 +1487,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1419,6 +1502,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 16,
       "estimatedLoadGb": 8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1437,6 +1521,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1451,6 +1536,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1465,6 +1551,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1479,6 +1566,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1493,6 +1581,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": false,
       "ggufDiy": false,
@@ -1507,6 +1596,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": false,
@@ -1521,6 +1611,7 @@
       "quantization": "API",
       "minRamGb": 0,
       "estimatedLoadGb": 0,
+      "kvKbPerToken": null,
       "runsLocally": false,
       "openWeights": true,
       "ggufDiy": true,
@@ -1535,6 +1626,7 @@
       "quantization": "Q6_K",
       "minRamGb": 32,
       "estimatedLoadGb": 24,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1553,6 +1645,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 32,
       "estimatedLoadGb": 23,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1571,6 +1664,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 24,
       "estimatedLoadGb": 19,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1589,6 +1683,7 @@
       "quantization": "MXFP4",
       "minRamGb": 96,
       "estimatedLoadGb": 65.4,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1607,6 +1702,7 @@
       "quantization": "MXFP4",
       "minRamGb": 24,
       "estimatedLoadGb": 13.8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1625,6 +1721,7 @@
       "quantization": "Q4_K_M",
       "minRamGb": 72,
       "estimatedLoadGb": 50.4,
+      "kvKbPerToken": 24,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1643,6 +1740,7 @@
       "quantization": "Q8_0",
       "minRamGb": 64,
       "estimatedLoadGb": 38.7,
+      "kvKbPerToken": 20,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1661,6 +1759,7 @@
       "quantization": "Q8_0",
       "minRamGb": 64,
       "estimatedLoadGb": 38.7,
+      "kvKbPerToken": 20,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1679,6 +1778,7 @@
       "quantization": "Q8_0",
       "minRamGb": 16,
       "estimatedLoadGb": 10.7,
+      "kvKbPerToken": 32,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1697,6 +1797,7 @@
       "quantization": "Q8_0",
       "minRamGb": 48,
       "estimatedLoadGb": 30,
+      "kvKbPerToken": 64,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1715,6 +1816,7 @@
       "quantization": "Q8_0",
       "minRamGb": 24,
       "estimatedLoadGb": 12.8,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1733,6 +1835,7 @@
       "quantization": "Q8_0",
       "minRamGb": 48,
       "estimatedLoadGb": 28.1,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1751,6 +1854,7 @@
       "quantization": "Q8_0",
       "minRamGb": 128,
       "estimatedLoadGb": 84.8,
+      "kvKbPerToken": 24,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1769,6 +1873,7 @@
       "quantization": "Q8_0",
       "minRamGb": 24,
       "estimatedLoadGb": 15.9,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1787,6 +1892,7 @@
       "quantization": "Q8_0",
       "minRamGb": 128,
       "estimatedLoadGb": 75,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,
@@ -1805,6 +1911,7 @@
       "quantization": "Q6_K",
       "minRamGb": 96,
       "estimatedLoadGb": 57.9,
+      "kvKbPerToken": null,
       "runsLocally": true,
       "openWeights": true,
       "ggufDiy": false,