Skip to main content
BenchLM

WeirdML v2 (WeirdML)

We show this table for reference; we do not rank on it.

A machine-learning engineering benchmark that tests whether LLMs can train models on novel datasets, write PyTorch code, and improve through iterative feedback.

Average accuracy on WeirdML — WeirdML v2

We mirror the published average accuracy view for WeirdML. GPT-5.5 leads the public snapshot at 84.91%, followed by GPT-5.5 (83.90%) and Claude Opus 4.8 (82.89%). We do not use these results to rank models overall.

25 modelsCodingCurrentDisplay onlyUpdated WeirdML v2

Average accuracy table (25 models)

Score
1
GPT-5.5OpenAI · Closed
84.91%
2
GPT-5.5OpenAI · Closed
83.90%
3
Claude Opus 4.8Anthropic · Closed
82.89%
4
Claude Opus 4.6Anthropic · Closed
77.95%
5
GPT-5.3 CodexOpenAI · Closed
77.90%
6
GPT-5.4OpenAI · Closed
77.70%
7
Claude Opus 4.7Anthropic · Closed
76.44%
8
Claude Opus 4.7Anthropic · Closed
76.40%
9
Claude Opus 4.8Anthropic · Closed
76.04%
10
Claude Opus 4.7Anthropic · Closed
75.45%
11
GPT-5.2OpenAI · Closed
72.19%
13
Claude Opus 4.8Anthropic · Closed
70.45%
15
GPT-5.5OpenAI · Closed
67.15%
16
Claude Sonnet 4.6Anthropic · Closed
66.07%
17
Claude Opus 4.6Anthropic · Closed
65.87%
18
Claude Opus 4.5Anthropic · Closed
63.74%
19
GPT-5.2OpenAI · Closed
63.44%
20
Gemini 3.5 FlashGoogle · Closed
62.64%
22
GPT-5.1OpenAI · Closed
60.77%
23
60.70%
24
60.39%
25
GPT-5.4 miniOpenAI · Closed
60.30%

How WeirdML is shown here

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.

Snapshot

25 mirrored top rows130 official CSV rows17 ML tasksOfficial CSVDisplay only

The published WeirdML snapshot places GPT-5.5 first at 84.91%. The third row is 2.02 points behind. The broader top-10 range is 9.46 points, so many of the published results sit in a relatively narrow band.

25 models have been evaluated on WeirdML. The benchmark falls in the Coding category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. WeirdML is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About WeirdML

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.

Freshness and provenance

Version

WeirdML 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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.

Questions

What does WeirdML measure?

A machine-learning engineering benchmark that tests whether LLMs can train models on novel datasets, write PyTorch code, and improve through iterative feedback.

Which model leads the published WeirdML snapshot?

GPT-5.5 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.

How many models are evaluated on WeirdML?

The WeirdML v2 snapshot contains 25 AI models.

Last updated: WeirdML v2 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.