reasoning benchmark
Internal Research Debugging Evaluation
The Internal Research Debugging Evaluation measures whether models can debug 41 real bugs from internal OpenAI research experiments (plus alignment-auditing tasks), where the original solutions took experienced researchers hours to days. Passing corresponds to providing assistance that would unblock the user, including partial root-cause explanations or fixes.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
Internal Research Debugging Evaluation leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | OP | 68.3% | 100.0% | 3 | C | |
| 2 | OP | 67.8% | 50.0% | 3 | C | |
| 3 | OP | 50.8% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
Internal Research Debugging Evaluation highlights
The top published results on this benchmark's own scale.
What is Internal Research Debugging Evaluation?
Definition and scoring fields from the benchmark registry.
The Internal Research Debugging Evaluation measures whether models can debug 41 real bugs from internal OpenAI research experiments (plus alignment-auditing tasks), where the original solutions took experienced researchers hours to days. Passing corresponds to providing assistance that would unblock the user, including partial root-cause explanations or fixes.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Internal Research Debugging Evaluation
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- internal-research-debugging-evaluation|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Internal Research Debugging Evaluation.
Which model scores highest on Internal Research Debugging Evaluation?
GPT-5.6 Sol is currently ranked first with 68.3%.
What does Internal Research Debugging Evaluation measure?
The Internal Research Debugging Evaluation measures whether models can debug 41 real bugs from internal OpenAI research experiments (plus alignment-auditing tasks), where the original solutions took experienced researchers hours to days. Passing corresponds to providing assistance that would unblock the user, including partial root-cause explanations or fixes.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.