{
  "title": "Yeti Claw Fleet Map: Where Each Inference Lane Fits",
  "published": "2026-05-07",
  "summary": {
    "mac_mini": {
      "role": "Primary small-surface text lane",
      "local_cost_per_hour_24x7": 0.110,
      "validated_concurrency": "4+",
      "representative_model": "qwen2.5:7b",
      "output_tokens_per_second": 47.77,
      "cost_per_million_output_tokens_24x7": 0.64,
      "recommended_use": "interactive text and low-drama public chat"
    },
    "dgx_spark": {
      "role": "Premium multimodal box for text plus image generation",
      "local_cost_per_hour_24x7": 0.254,
      "interactive_text_band": "2 premium / 4 shared / 8 stable but queued",
      "representative_text_model": "qwen3:8b",
      "output_tokens_per_second": 41.97,
      "cost_per_million_output_tokens_24x7": 1.68,
      "image_sweep_success_rate": "8/10",
      "recommended_use": "image generation, multimodal development, and premium shared text"
    },
    "chewbacuh": {
      "role": "Fast BeastMode text lane",
      "vm_shape": "8 vCPU / 48 GiB",
      "throughput_rps_plateau": 0.098,
      "baseline_latency_seconds": 10.37,
      "lower_bound_cost_per_million_output_tokens_24x7": 1.04,
      "recommended_use": "supplemental public text capacity where queueing is acceptable"
    },
    "lil_beastly": {
      "role": "Heavier BeastMode text lane",
      "vm_shape": "12 vCPU / 96 GiB",
      "throughput_rps_plateau": 0.051,
      "baseline_latency_seconds": 21.20,
      "lower_bound_cost_per_million_output_tokens_24x7": 3.98,
      "recommended_use": "larger-model availability rather than cheapest real-time output"
    }
  },
  "routing_guidance": [
    {
      "scenario": "Public interactive text",
      "preferred_lane": "Mac mini",
      "fallback_lane": "Chewbacuh",
      "reason": "best latency-to-cost balance"
    },
    {
      "scenario": "Premium multimodal or image generation",
      "preferred_lane": "DGX Spark",
      "fallback_lane": "queue job on Spark",
      "reason": "only lane with proven image-generation coverage"
    },
    {
      "scenario": "Larger-model public text",
      "preferred_lane": "LiL-Beastly",
      "fallback_lane": "Chewbacuh",
      "reason": "14B availability with known queue-heavy behavior"
    }
  ],
  "source_reports": [
    "/mission-control/cloud-vs-local-inference-economics-may-2026/",
    "/mission-control/beastmode-inference-lanes-may-2026/",
    "/mission-control/text-capacity-review-may-2026/",
    "/mission-control/spark-image-model-sweep-april-2026/"
  ]
}