Håvard Ihle

@htihle

Benchmarks @EpochAIResearch, I made WeirdML, worried about ASI, views are my own.

following
followers

GPT-6.1 Terra launches with 50% lower standard API prices than GPT-5.6 Terra

OpenAI says Terra can interpret drawings, act on application screens and draft text.

117 Sources, 13d ago, updated 13d ago

YouTube

Gemini 4 Argon (high) reportedly scores 20.2% on WeirdML v3

The benchmark's creator says that puts it between Sonnet 5.5 and Fable 5.1, but behind the frontier on token efficiency.

2 Sources, 24d ago, first seen 24d ago

Muse Wonder (xhigh) reportedly scores 14.9% on WeirdML v3

A user says its score was close to GPT 5.6 Sol and Fable 5 in a Codex CLI harness run.

2 Sources, 26d ago, first seen 26d ago

WeirdML v3 compares GPT 6.1 Sol, Sonnet 5.5 and Grok 4.7, with some results incomplete

A post sharing the results says Sonnet 5.5 scores better than Opus 5, while Grok 4.7 is ahead of Kimi-K3.

3 Sources, 40d ago, updated 40d ago

WeirdML v3 debuts with 11 hand-made tasks for AI models

The announcement says models must produce results despite limited data, unspecified goals and/or very limited feedback.

5 Sources, 54d ago, first seen 54d ago

GPT 6 Astra (high) ties WeirdML’s top score at 92.9%, a user reports

The user reports new individual high scores on 6 of 17 tasks and significantly less code than Sol and other recent GPT models, though still more than Claude.

1 Source, 67d ago, first seen 67d ago

Claude Fable 5.1 (high) scores 92.3% on WeridML, a user reports

The user reports a new benchmark high, beating Fable 5 (max) by 0.4%, with new records on 3 of the 17 tasks.

1 Source, 69d ago, first seen 69d ago

Featured in

GPT-6.1 Terra launches with 50% lower standard API prices than GPT-5.6 Terra

Gemini 4 Argon (high) reportedly scores 20.2% on WeirdML v3

Muse Wonder (xhigh) reportedly scores 14.9% on WeirdML v3

WeirdML v3 compares GPT 6.1 Sol, Sonnet 5.5 and Grok 4.7, with some results incomplete

WeirdML v3 debuts with 11 hand-made tasks for AI models

GPT 6 Astra (high) ties WeirdML’s top score at 92.9%, a user reports

Claude Fable 5.1 (high) scores 92.3% on WeridML, a user reports

Featured in

GPT-6.1 Terra launches with 50% lower standard API prices than GPT-5.6 Terra

Gemini 4 Argon (high) reportedly scores 20.2% on WeirdML v3

Muse Wonder (xhigh) reportedly scores 14.9% on WeirdML v3

WeirdML v3 compares GPT 6.1 Sol, Sonnet 5.5 and Grok 4.7, with some results incomplete

WeirdML v3 debuts with 11 hand-made tasks for AI models

GPT 6 Astra (high) ties WeirdML’s top score at 92.9%, a user reports

Claude Fable 5.1 (high) scores 92.3% on WeridML, a user reports