Search-SWE
GPU TASK Valid result

Query-side Encoder Alignment

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

Task score (%)28.9
Input Tokens11,669,767including 11,453,733 cached
Output Tokens143,052
Agent time46min
Cost$0.21
Recorded steps104

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
unspecified_in_legacy_metadata
Coding started
2026-10-23T18:43:07.196011Z
Coding finished
2026-10-23T19:29:51.914642Z
Input (includes cache)
11,669,767
Cached input (subset)
11,453,733
Cache writes (subset)
215,731
Output
143,052
Reasoning (reported separately; not added)
113,214
Recorded responses
101

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": "official_harbor_regrade",
  "task_commit": "1d73c6f9c4a90ba922b7572d9395a6f40661bbb2",
  "verifier_commit": "1d73c6f9c4a90ba922b7572d9395a6f40661bbb2",
  "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
query-encoder-alignment-gpt-6-luna-max-e135b4cc14bc
Source trajectory SHA-256
9ccb82511f8242430a3b0408356d491810e0fef3ce1b05abfb5b8bbfb6a077c0
Final reward SHA-256
345ca680b21d40ac72f581e1968cd90ec03ac8063b8368524bc16c0f6e93a60b