GPT-6.1 Terra scores 28.1% on agentic WeirdML v3, matching Claude Fable 5.1 at lower cost
scaling01 · x · 2026-10-29
WeirdML v3 has launched: a fully agentic benchmark with 11 hand-made complex tasks where models must explore unfamiliar data and build ML/data-analysis pipelines under unspecified goals and very limited feedback.
Latest results show GPT-6.1 Terra (xhigh) at 28.1%, comparable to Claude Fable 5.1: Terra is more token-efficient while Fable achieves a higher final score. The chart shows Terra's official score at 28.1%, final best 40.4%, cost per run $2.86, running on Codex CLI v0.164.0 across 98 valid runs. The author notes Terra is much cheaper than any comparable models.
More from Models
- DeepMind employees unusually confident as Google launches its Carbon model — bookwormengr · 2026-10-29
- Alphabet stock dumps 8% on earnings miss as analysts debate AI capex payback — MatthewBerman · 2026-10-29
- Fable 5 praised for blending deep scientific knowledge with storytelling delight — repligate · 2026-10-29
- ChatGPT's unified UI confuses users: Pro model hides under GPT-6 Sol, not Astra — Angaisb_ · 2026-10-29
- User burns through OpenAI's $100 weekly quota in under 10 hours — teortaxesTex · 2026-10-29
- Hy-Nomas interactive video world model demo launches on Hugging Face — _akhaliq · 2026-10-29