Search-SWE
CPU TASK Valid result

Long-PDF Sentence Evidence — Vectorless

GPT-6 Luna · Codex (v0.164.0) · reasoning effort max

Task score (%)8.3
Input Tokens24,541,032including 24,138,950 cached
Output Tokens246,671
Agent time1h 9min
Cost$0.59
Recorded steps119

Cost is estimated from recorded agent tokens at official API rates in USD (2026-11-01; DeepSeek off-peak pricing).

Loading trajectory…

Select this tab to browse the saved files.

Coding time & token usage
Original agent completion
normal
Coding started
2026-10-24T13:44:03.564361Z
Coding finished
2026-10-24T14:53:53.644351Z
Input (includes cache)
24,541,032
Cached input (subset)
24,138,950
Cache writes (subset)
366,212
Output
246,671
Reasoning (reported separately; not added)
190,248
Recorded responses
116

Input includes cached input; output is counted once. Auxiliary reasoning counts are shown as recorded. Regrading does not repeat coding usage.

Select this tab to browse the saved files.

Configuration

Evaluation settings summary
{
  "agent": "Codex",
  "agent_timeout_sec": null,
  "coding_commit": null,
  "effort": "max",
  "model": "openai/gpt-6-luna",
  "provider": "OpenRouter",
  "resources": null,
  "run_mode": "saved_evidence_uniform_semantic_regrade",
  "task_commit": "7e7ce5c052aa4d0cace60dc9a1c419095d521c7c",
  "verifier_commit": "7e7ce5c052aa4d0cace60dc9a1c419095d521c7c",
  "version": "0.164.0"
}

Select this tab to browse the saved files.

Resolved configuration

Select this tab to browse the saved files.

Trial log

Select this tab to browse the saved files.

Exceptions

Select this tab to browse the saved files.

Select this tab to browse the saved files.

Trajectory downloads & provenance

Public text redacts private values and normalizes host paths. Source hashes identify the original records.

Trial ID
long-pdf-evidence-vectorless-gpt-6-luna-max-eaa6aab1e7fd
Source trajectory SHA-256
ab86027e21f26d81c6c4da6124f7cbcf3f1bb3fdc00cd0af21e723b586fa7568
Final reward SHA-256
6576ae69a703e3219da2a8380a23846a562d086db80abc46bd9d27c966bf1a79