{
  "schema": "runaii-agent-workload/1",
  "measured_at": "2026-10-24",
  "source": "runaii.cloud gateway aggregate; the same figures published at https://runaii.cloud/llms.txt",
  "unit": "prompt tokens (input), plus output tokens where stated",
  "workload": {
    "all_time_prompt_tokens": 1600000000,
    "peak_day_prompt_tokens": 224000000,
    "cache_read_share_of_prompt": 0.93
  },
  "derived": {
    "peak_day_cached_tokens": 208320000,
    "peak_day_fresh_tokens": 15680000,
    "assumed_output_tokens_per_peak_day": 8000000,
    "modelled_task_shape": {
      "fresh_input_tokens": 50000,
      "cached_prompt_tokens": 200000,
      "output_tokens": 8000,
      "total_tokens": 258000
    },
    "tasks_per_peak_day_at_modelled_shape": 868,
    "peak_day_cost_usd": {
      "deepseek-v4-flash": 13.31,
      "qwen3.8-flash": 27.4,
      "glm-5.3": 111.32
    },
    "thirty_peak_days_cost_usd": {
      "deepseek-v4-flash": 399,
      "qwen3.8-flash": 822,
      "glm-5.3": 3339
    },
    "peak_day_share_of_all_time_tokens": 0.14
  },
  "rates_used": {
    "deepseek-v4-flash": { "fresh_per_1m": 0.22, "cached_per_1m": 0.022, "output_per_1m": 0.66 },
    "qwen3.8-flash": { "fresh_per_1m": 0.4, "cached_per_1m": 0.04, "output_per_1m": 1.6 },
    "glm-5.3": { "fresh_per_1m": 1.4, "cached_per_1m": 0.26, "output_per_1m": 4.4 }
  },
  "not_published": [
    "per-day series (the aggregate is the receipt, the daily curve is not)",
    "per-model token split",
    "per-customer or per-key usage",
    "cache-hit distribution (the 93% is a share of prompt volume, not a median)",
    "measured throughput, latency or benchmark results"
  ],
  "notes": [
    "Output tokens per peak day (8M) is a modelling assumption carried over from the published coding-agent cost comparison; it is not a measured figure.",
    "The 258K-token task shape is a modelling choice, stated so it can be re-priced against any rate card.",
    "All cost figures are arithmetic on published rates from src/lib/catalog.ts on 2026-10-24, not invoices."
  ]
}