{
  "status": "ok",
  "verdict": "self_host_needs_throughput",
  "headline": "Estimate. The API is cheaper below 457 million tokens a day. Above that, self-hosting wins only if the configuration is busy at least 12% of the time at our highest throughput estimate, and 69% at our lowest.",
  "headline_kind": "busy_at_least_between_estimates",
  "headline_values": {
    "prefix": "estimate",
    "break_even_m": 457,
    "busy_percent_low": 12,
    "busy_percent_high": 69
  },
  "confidence": "derived",
  "threshold_m_tokens_per_day": 457.14,
  "required_tokens_per_second": {
    "at_full_utilization": 5291.1,
    "at_utilization": 10582.1,
    "utilization_percent": 50
  },
  "break_even_utilization_percent": {
    "prudent": 68.6,
    "median": 27.5,
    "optimistic": 11.8
  },
  "utilization_percent": 50,
  "api": {
    "provider": "Thinking Machines",
    "usd_per_m_tokens": 0.525,
    "checked_at": "2026-10-31T16:45:58.753Z",
    "offers_compared": 3,
    "mix": {
      "source": "default_3_1",
      "input_share_percent": 75
    },
    "basis": "cheapest of 3 current standard offers for a 3:1 input:output mix (blended), promotions, unstable prices, off-peak prices without published hours and unverified prices far below those of the other providers excluded"
  },
  "model": {
    "id": "thinkingmachines/inkling-small",
    "name": "Thinking Machines: Inkling Small",
    "params_b": 265.956,
    "kind": "moe",
    "active_params_b": 12,
    "weights_gb": 265.96,
    "quantization": "fp8"
  },
  "configuration": {
    "vendor": "NVIDIA",
    "throughput_evidence": "derived",
    "gpu_id": "l40-48gb",
    "gpu_name": "L40 48GB",
    "gpus_needed": 8,
    "gpus_billed": 8,
    "provider": "CoreWeave",
    "region": "north-america",
    "tier": "guaranteed",
    "usd_per_gpu_hour": 1.25,
    "usd_per_hour": 10,
    "usd_per_day": 240,
    "vram_total_gb": 384,
    "weights_share": 0.693,
    "checked_at": "2026-10-31T18:22:18.044Z",
    "source_url": "https://www.coreweave.com/pricing"
  },
  "weights_margin": {
    "used_percent": 70,
    "default_percent": 70,
    "explanation": "The weights may fill 70% of the GPU memory; the rest is the engine's reserve and the context cache (default 70%)."
  },
  "throughput": {
    "status": "ok",
    "confidence": "derived",
    "method": "same architecture type, scaled by memory bandwidth and active weights (x1.92)",
    "tokens_basis": "output",
    "reference": {
      "model_id": "minimax/minimax-m3",
      "gpu_id": "b200-sxm-180gb",
      "source_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/28604484366/attempts/2"
    }
  },
  "scenarios": [
    {
      "name": "prudent",
      "tokens_per_second": 7708.8,
      "raw_capacity_m_tokens_per_day": 666.04,
      "useful_capacity_m_tokens_per_day": 333.02,
      "break_even_utilization_percent": 68.6,
      "cost_per_m_tokens": 0.7207,
      "vs_api": "dearer"
    },
    {
      "name": "median",
      "tokens_per_second": 19271.9,
      "raw_capacity_m_tokens_per_day": 1665.09,
      "useful_capacity_m_tokens_per_day": 832.55,
      "break_even_utilization_percent": 27.5,
      "cost_per_m_tokens": 0.2883,
      "vs_api": "cheaper"
    },
    {
      "name": "optimistic",
      "tokens_per_second": 44851,
      "raw_capacity_m_tokens_per_day": 3875.13,
      "useful_capacity_m_tokens_per_day": 1937.56,
      "break_even_utilization_percent": 11.8,
      "cost_per_m_tokens": 0.1239,
      "vs_api": "cheaper"
    }
  ],
  "volume_m_tokens_per_day": null,
  "assumptions": [
    "Hardware rental only: the hourly price the provider publishes for the machine. It does not count the engineering to deploy and run the model, monitoring, redundancy, start-up time, or storage and network costs billed separately.",
    "Configuration: we pick the cheapest one whose weights fit. If a configuration costing up to 25% more has a measured throughput, we pick it instead; failing that, one with a derived throughput; failing that, one with a published lower bound that settles the verdict; among equals, the cheaper. A configuration whose throughput is only a wide estimate is never preferred over the cheapest. This is our rule, not a technical requirement.",
    "Quantization fp8; guaranteed tier prices as published by the provider; utilization 50%.",
    "Published figures are generally peak throughput measured at high concurrency; an interactive service with latency targets sustains less.",
    "Throughput is aggregate (input and output tokens, 3:1). Published figures in generated tokens are converted with a prefill/generation speed ratio of 3 (prudent), 5 (median) and 8 (optimistic). We treat a published output figure as if the GPUs only generated tokens, ignoring that the benchmark also read prompts: this understates the throughput of the whole configuration, which is the cautious side for self-hosting.",
    "Break-even = daily GPU cost / API price per million tokens (457 million tokens a day). Self-hosting is cheaper only if the average utilization you can sustain exceeds break-even / raw capacity (the break-even utilization of each scenario), whatever the number of GPUs."
  ],
  "fetched_at": "2026-10-31T18:33:37.292Z"
}