@charliejhills: A hidden model just beat Claude's best. Everyone thinks it's Fable 5.5. Here's what we know: It's called Obsidian 3. Cognition's FrontierCode 2.0 data had it. It scored 46.8%. Opus 5.5 scored 45.5%. It ran in Claude Code, like every Claude model. But I looked at the chart. Every Claude model on it is orange. Obsidian 3 is blue, the same blue as Meta's model. You'd pay about 3.6x Opus at the same effort. At max effort, both scored 45.5%. A day later, someone pulled the row. Anthropic hasn't announced Fable 5.5. So I'm staying on Opus 5.5 for now. You can use it today, and it's basically level. Would you pay nearly 4x for the same score?

Charlie Hills

Open In TikTok:

Region: GB

Monday 02 November 2026 09:18:23 GMT