LLM Stats (Internal Research Debugging Evaluation): leaderboard

OpenAI's Internal Research Debugging evaluation, as tracked by LLM Stats: models receive code and logs from failed internal OpenAI research experiments (41 real bugs that took researchers hours to days to debug, plus six alignment-auditing tasks) and are scored by mean rubric credit (%) for diagnosing root causes and resolutions, including partial credit and counting missing scores as zero. Higher is better. LLM Stats combines GPT-6.2 Astra and Sol scores from the GPT-6.2 system card (Table 5) with GPT-5.6 Sol, Terra and Luna scores from the GPT-5.6 launch post.

Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around February 2027. 5 models tracked. Updated 19 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6.2 Astra81.85#4
2GPT-6.2 Sol74.66#12
3GPT-5.6 Sol68.3#36
4GPT-5.6 Terra67.8#72
5GPT-5.6 Luna50.8#133

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #5 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=llm-stats-internal-research-debugging-evaluation · How It Works · Data refreshed daily, snapshot 2026-11-20.