Search-SWE
GPU TASK Valid result

Training Error Tracing

DeepSeek V4.1 Flash 1017 · Codex (v0.164.0) · reasoning effort max

Task score (%)0.0
Input Tokens103,296,442including 99,769,088 cached
Output Tokens330,365
Agent time2h 51min
Cost$1.03
Recorded steps491

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
2026-10-23T18:50:15.702799Z
Coding finished
2026-10-23T21:41:25.538174Z
Input (includes cache)
103,296,442
Cached input (subset)
99,769,088
Cache writes (subset)
0
Output
330,365
Reasoning (reported separately; not added)
213,383
Recorded responses
488

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": 14400.0,
  "coding_commit": null,
  "effort": "max",
  "model": "deepseek/deepseek-v4.1-flash-1017",
  "provider": "OpenRouter",
  "resources": null,
  "run_mode": "official_harbor_regrade",
  "task_commit": null,
  "verifier_commit": "8287076ba397b9266c9a399a857a89ba4815e2cb",
  "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
training-error-tracing-deepseek-v4.1-flash-1017-max-2419e1ce895c
Source trajectory SHA-256
ee524771980cbb7f01c3f3ef212525bb5aaa5851cb5f034e3606a8d1fa5c61e7
Final reward SHA-256
4c105b06cff24c88ff19850fa179d69e9459fbc88285320493c27ef47952d2ba