{
  "as_of": "2026-10-24",
  "model": "gpt-6.1-sol",
  "tier": "standard, short-context pricing (up to 272K input tokens)",
  "source": "https://developers.openai.com/api/docs/pricing",
  "rules_source": "https://developers.openai.com/api/docs/guides/prompt-caching#cache-lifetime",
  "prefix_tokens": 100000,
  "usd_per_million": {
    "uncached": 2.0,
    "write_30m": 2.5,
    "write_12h": 5.0,
    "read": 0.1
  },
  "scenarios": [
    {"name": "Chatty session, turns 3 min apart", "every_minutes": 3, "count": 20},
    {"name": "Agent waits 2 h for tests, six times", "hours": [0, 2, 4, 6, 8, 10]},
    {"name": "Returns every 45 min through a working day", "every_minutes": 45, "count": 13},
    {"name": "Codex-style: three quick turns, back after ~4 h", "hours": [0, 0.1667, 0.3333, 4.3333, 4.5]},
    {"name": "Three turns, then back the next morning (15 h)", "hours": [0, 0.25, 0.5, 15.5, 15.75, 16]},
    {"name": "Daily job, 24 h apart", "hours": [0, 24, 48, 72, 96]},
    {"name": "Branch cancelled after one call", "hours": [0]}
  ],
  "assumptions": [
    "One identical 100000-token prefix on each request; request stays in the standard short-context price band.",
    "30m: a write or read gives coverage until at least 30 minutes after it (documented refresh rule).",
    "12h: coverage lasts at least 12 hours after the write. Whether a 12h read restarts the clock is undocumented; both variants are computed and give identical results for these scenarios.",
    "Every request inside documented coverage is a full-prefix hit; no reuse after the minimum (conservative, not documented eviction).",
    "No-cache comparator makes no cache writes and receives no cached pricing.",
    "No prefix growth, maintenance requests, output, tool costs, retries or other application costs.",
    "No production cache behavior, infrastructure cost or latency was measured."
  ],
  "timeline": {
    "long_write_hour": 0,
    "short_read_hour": 10,
    "long_minimum_hours": 12,
    "short_minimum_hours": 0.5,
    "upgrade_hour": 0.25
  }
}