Fable 5.2 tested: max effort costs 25x more than medium for just +0.06 score
aakashgupta · x · 2026-11-11
Aakash Gupta blind-tested Fable 5.2 across eight product-management tasks: max effort scored 9.07/10 after 12 minutes of thinking at $3.27/task, versus 9.01 after 30 seconds at $0.13 — the extra $3.14 bought 0.06 points. His deep-dive also tests Spotify's viral Claude Code token-saving trick (its co-CEO said most senior engineers haven't written code since December) to verify the claimed up-to-90% savings. This week Anthropic reclaimed the Artificial Analysis Intelligence Index top spot with Fable 5.2 at 65, ending Google's 12-day lead.
Related event: Test Shows 12 Extra Minutes of Model Thinking Costs 25x for 0.06 Point Gain(2 posts)→
More from coding & agent
- Grok Bots can now make phone calls over iMessage — 'zero learning curve is the sauce' — RileyRalmuto · 2026-11-11
- Open-source artificial life simulator ALIEN 5.0 beta lets an AI agent take over evolution via MCP on one RTX 4090 — BertChakovsky · 2026-11-11
- Hackathon organizer proposes two-round peer judging as AI inflates viable submissions — schwentker · 2026-11-11
- Google Antigravity announces /swarm for parallel multi-agent orchestration — algo_diver · 2026-11-11
- Agency posts detailed wishlist for hiring agentic AI project management buildout — Accomplished-Dust185 · 2026-11-11
- SF developer meetup 'Who's Watching the Agents?' to cover agent observability — davidcrawshaw · 2026-11-11