A machine-learning engineering benchmark that tests whether LLMs can train models on novel datasets, write PyTorch code, and improve through iterative feedback.
BenchLM mirrors the top public WeirdML v2 rows from the official CSV. WeirdML tests whether models can do machine-learning engineering on 17 novel datasets, reporting average accuracy across tasks.
WeirdML is display only on BenchLM because the task is a specialized ML-agent evaluation with iterative feedback and execution constraints. BenchLM stores it as context rather than a weighted model-only benchmark.
BenchLM mirrors the published average accuracy view for WeirdML. gpt-5.5 (xhigh) leads the public snapshot at 84.91% , followed by gpt-5.5 (high) (83.90%) and claude-opus-4.8 (xhigh) (82.89%). BenchLM does not use these results to rank models overall.
gpt-5.5 (xhigh)
OpenAI
gpt-5.5:xhigh
gpt-5.5 (high)
OpenAI
gpt-5.5
claude-opus-4.8 (xhigh)
Anthropic
claude-opus-4.8:xhigh
The published WeirdML snapshot is tightly clustered at the top: gpt-5.5 (xhigh) sits at 84.91%, while the third row is only 2.02 points behind. The broader top-10 spread is 9.46 points, so many of the published scores sit in a relatively narrow band.
25 models have been evaluated on WeirdML. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. WeirdML is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
Year
2026
Tasks
17 novel ML engineering tasks
Format
Average accuracy across tasks
Difficulty
Novel dataset modeling and iterative debugging
WeirdML v2 evaluates models on 17 unusual ML tasks and reports average accuracy across tasks from the official CSV. BenchLM mirrors the top official rows as display-only ML-agent evidence.
Version
WeirdML 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
A machine-learning engineering benchmark that tests whether LLMs can train models on novel datasets, write PyTorch code, and improve through iterative feedback.
gpt-5.5 (xhigh) currently leads the published WeirdML snapshot with 84.91% average accuracy. BenchLM shows this benchmark for display only and does not use it in overall rankings.
25 AI models are included in BenchLM's mirrored WeirdML snapshot, based on the public leaderboard captured on WeirdML v2.
For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.