{
  "scope": "Tokens and runtime for solver inference only, over the 68 finished runs of the sweep. Judging and local infrastructure are excluded. No cost is reported: see the cost field.",
  "cost": "unavailable",
  "cost_reason": "Cognition publishes no per-token rate for SWE-2. Its catalog cost tier Free is a promotion dated through 2026-10-10, not a rate, so any figure derived from it would read as a measured price and would stop being true when the promotion ends. The population record carries a null cost for every run. What is measured instead is tokens, runtime and Devin's own credit and ACU counters, which read zero throughout.",
  "cost_revisit": "If Cognition publishes a per-token API rate for SWE-2, these runs can be repriced from their receipts and the column stops being unavailable.",
  "sources": {
    "devin": "Devin CLI receipts (the per-request usage records in the CLI's agent stream)"
  },
  "rates_per_million": null,
  "pricing_verified": null,
  "groups": [
    {
      "model": "swe2",
      "effort": "medium",
      "n": 23,
      "runs_attempted": 23,
      "usd": null,
      "usd_total": null,
      "cost": "unavailable",
      "raw_tokens": {
        "n": 23,
        "mean": 16802267.652173914,
        "se": 4729670.587117841
      },
      "raw_tokens_total": 386452156,
      "output_tokens": {
        "n": 23,
        "mean": 220480.60869565216,
        "se": 53092.46917063976
      },
      "output_tokens_total": 5071054,
      "minutes": {
        "n": 23,
        "mean": 49.778971014492754,
        "se": 9.590646328891255
      },
      "solver_fallback_runs": 0
    },
    {
      "model": "swe2",
      "effort": "high",
      "n": 22,
      "runs_attempted": 23,
      "usd": null,
      "usd_total": null,
      "cost": "unavailable",
      "raw_tokens": {
        "n": 22,
        "mean": 7526272.954545454,
        "se": 944121.6128164232
      },
      "raw_tokens_total": 165578005,
      "output_tokens": {
        "n": 22,
        "mean": 136664.68181818182,
        "se": 16406.226969080362
      },
      "output_tokens_total": 3006623,
      "minutes": {
        "n": 22,
        "mean": 35.08740681818182,
        "se": 4.571588221704351
      },
      "solver_fallback_runs": 0
    },
    {
      "model": "swe2",
      "effort": "max",
      "n": 23,
      "runs_attempted": 23,
      "usd": null,
      "usd_total": null,
      "cost": "unavailable",
      "raw_tokens": {
        "n": 23,
        "mean": 9730538,
        "se": 1185163.8018067444
      },
      "raw_tokens_total": 223802374,
      "output_tokens": {
        "n": 23,
        "mean": 170843.13043478262,
        "se": 16394.985984984654
      },
      "output_tokens_total": 3929392,
      "minutes": {
        "n": 23,
        "mean": 56.64182536231884,
        "se": 6.4667329492261585
      },
      "solver_fallback_runs": 0
    }
  ],
  "totals": {
    "swe2": {
      "runs_attempted": 69,
      "runs": 68,
      "usd": null,
      "raw_tokens": 775832535,
      "output_tokens": 12007069,
      "solver_hours": 53.660021111111114
    }
  },
  "limitations": [
    "SWE-2 has no public API price, so no cost is estimated and the cost column reads unavailable; the sweep ran on a Devin subscription. The catalog's Free tier is a promotion dated through 2026-10-10, not a published rate.",
    "Tokens are the Devin CLI's per-request usage receipts, deduplicated by request id: uncached input, cache reads and output.",
    "Devin's own credit and ACU counters are recorded from the receipts and were zero for every run of this sweep. A zero counter on a subscription is not a price, so it is reported as a counter and not as a cost.",
    "The 68 finished runs are all counted here, the three that changed no source file included. The one run that did not finish has no receipt and is listed separately; its 3.0 hours of wall clock are not in these totals.",
    "Runtime is solver wall clock as the harness recorded it, judging excluded."
  ],
  "comparison_sha256": "9473e196c55443be8daab9cd1cbdd97597dc124bf6903df2924589d86209bb68",
  "comparison_judged_sha256": "e93d351eb4a2257d015c3e329ed6ecacc9b95b2e436c2a3df279b90f621c6f31",
  "note": "Per-run records are in runs.json (raw_tokens, token_usage, devin_credits, devin_acu; estimated_usd is null throughout). Judging is excluded."
}