{"name":"diego.technology Local AI Index","version":"1.0","date":"2026-10-02","hardware":"GMKtec EVO-X2 AI · AMD Ryzen AI Max+ 395 · 128 GB LPDDR5X-8000","weights":{"capability":50,"hard":20,"speed":15,"consistency":15},"ranking":[{"rank":1,"model":"gpt-oss-120b","quant":"MXFP4","status":"current","score":98.4,"parts":{"capability":49.7,"hard":19.8,"speed":14.9,"consistency":14.1},"runs":3,"genTps":48.5},{"rank":2,"model":"Gemma 4 31B QAT","quant":"UD-Q4_K_XL","status":"current","score":94.4,"parts":{"capability":50,"hard":20,"speed":9.4,"consistency":15},"runs":3,"genTps":10.6},{"rank":3,"model":"Qwen3-Coder-Next","quant":"UD-Q4_K_XL","status":"contender","score":91.1,"parts":{"capability":46.6,"hard":17.6,"speed":14.8,"consistency":12.2},"runs":3,"genTps":46.8},{"rank":4,"model":"GLM-5.3-Flash","quant":"UD-IQ3_XXS","status":"contender","score":89.2,"parts":{"capability":50,"hard":20,"speed":10.2,"consistency":9},"runs":1,"genTps":13.4},{"rank":5,"model":"KAT-Coder v2.5","quant":"Q8_0","status":"retired","score":83,"parts":{"capability":43.9,"hard":15.1,"speed":15,"consistency":9},"runs":1,"genTps":51.4},{"rank":6,"model":"Qwen3-Coder-Next","quant":"Q6","status":"retired","score":83,"parts":{"capability":43.4,"hard":16.3,"speed":14.3,"consistency":9},"runs":1,"genTps":41},{"rank":7,"model":"Muse Glimmer 30B","quant":"Q8_0","status":"contender","score":82.7,"parts":{"capability":47.5,"hard":18,"speed":8.2,"consistency":9},"runs":1,"genTps":7.6},{"rank":8,"model":"Tiel-Coder 35B","quant":"Q8_0","status":"retired","score":81.4,"parts":{"capability":42.3,"hard":15.6,"speed":14.5,"consistency":9},"runs":1,"genTps":43.6},{"rank":9,"model":"Nemotron 3 Super 120B","quant":"UD-Q4_K_XL","status":"contender","score":81,"parts":{"capability":45.8,"hard":16.7,"speed":11,"consistency":7.5},"runs":3,"genTps":16.8},{"rank":10,"model":"DeepSeek V4-Flash","quant":"UD-IQ3_XXS","status":"contender","score":79.7,"parts":{"capability":45,"hard":16,"speed":9.7,"consistency":9},"runs":1,"genTps":11.8},{"rank":11,"model":"Qwen3.8-Flash-Next","quant":"IQ4_XS-PLE · fixed sampling","status":"retired","score":78.4,"parts":{"capability":42.5,"hard":14,"speed":12.9,"consistency":9},"runs":1,"genTps":28.2},{"rank":12,"model":"Qwen3.8-Flash-Next","quant":"UD-IQ4_XS · fixed sampling","status":"retired","score":77.7,"parts":{"capability":42.5,"hard":14,"speed":12.2,"consistency":9},"runs":1,"genTps":23.6},{"rank":13,"model":"Gemma 4 26B-A4B QAT","quant":"UD-Q4_K_XL","status":"retired","score":71,"parts":{"capability":35,"hard":12,"speed":15,"consistency":9},"runs":1,"genTps":57.5},{"rank":14,"model":"Devstral 2 24B","quant":"Q8_0","status":"retired","score":70,"parts":{"capability":36.9,"hard":15.5,"speed":8.6,"consistency":9},"runs":1,"genTps":8.5},{"rank":15,"model":"Qwen3.8-27B","quant":"Q8_0","status":"audit","score":66,"parts":{"capability":36.9,"hard":12,"speed":8.2,"consistency":9},"runs":1,"genTps":7.5},{"rank":16,"model":"Gemma 4 12B QAT","quant":"UD-Q4_K_XL","status":"retired","score":62,"parts":{"capability":32.5,"hard":8,"speed":12.5,"consistency":9},"runs":1,"genTps":25.5},{"rank":17,"model":"Mistral Small 4 119B","quant":"UD-Q4_K_XL","status":"retired","score":61.7,"parts":{"capability":28.6,"hard":10.3,"speed":13.8,"consistency":9},"runs":1,"genTps":36.1}],"radar":[{"name":"Nex-N2.5-mini","maker":"—","size":"—","status":"downloaded","why":"Integrity-checked download complete. Compatibility smoke test next.","source":"https://huggingface.co/models?search=Nex-N2.5"},{"name":"Occamy 1.0","maker":"—","size":"—","status":"downloaded","why":"Integrity-checked download complete. Compatibility smoke test next.","source":"https://huggingface.co/models?search=Occamy"},{"name":"MiMo-V2.6-Distill-Qwen-9B","maker":"—","size":"9B (per name)","status":"downloaded","why":"A small distill of the current top open-weight family. Smoke test next.","source":"https://huggingface.co/models?search=MiMo-V2.6-Distill"},{"name":"Ling-3.0-flash-Fin","maker":"—","size":"—","status":"downloaded","why":"Integrity-checked download complete. Compatibility smoke test next.","source":"https://huggingface.co/models?search=Ling-3.0-flash"},{"name":"Ornith-1.5-35B-A3B","maker":"Ornith AI","size":"36B total · 3B active","status":"shortlist","why":"Agentic-coding MoE with publisher-reported SWE-bench Verified of 79%. Small active size should be fast here.","source":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B"},{"name":"Xing4.0-29B-A4B","maker":"China Telecom AI","size":"29B total · 4B active","status":"shortlist","why":"Agent-oriented MoE, Apache 2.0, 256K context; publisher reports 75 on SWE-bench Verified.","source":"https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B"},{"name":"DeepSeek-V4.1-Flash","maker":"DeepSeek","size":"Check fit","status":"shortlist","why":"Successor to V4-Flash, which scored 18/20 here at IQ3. Needs a quant that fits 96 GB of GPU memory.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash"},{"name":"MiMo-V2.6-Flash","maker":"Xiaomi","size":"309B total · 15B active","status":"shortlist","why":"Strong agent scores, MIT. Only fits at roughly 2–2.5 bits per weight — a stress test of aggressive quantization.","source":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL"},{"name":"Ternary Bonsai 2 27B","maker":"Prism ML","size":"27B at ~1.75 bits · 6 GB","status":"blocked","why":"A ternary Qwen3.8-27B that claims 98% of full quality. Needs a custom llama.cpp fork with CUDA, Metal or CPU kernels — no AMD GPU path yet.","source":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf"},{"name":"MiMo-V2.6-Pro","maker":"Xiaomi","size":"1.02T total · 42B active","status":"too-big","why":"Top of the public open-weight rankings, but about four times too big for 128 GB at any usable quant.","source":"https://huggingface.co/XiaomiMiMo"},{"name":"Kimi K3","maker":"Moonshot AI","size":"2.8T total","status":"too-big","why":"Far beyond a single 128 GB machine.","source":"https://huggingface.co/moonshotai"},{"name":"GLM-5.3","maker":"Z.ai","size":"~750B total (reported)","status":"too-big","why":"The full model; its Flash sibling already scored 20/20 here.","source":"https://huggingface.co/zai-org/GLM-5.3"}]}