SemiAnalysis: Claude Fable 5.2 spends 4.3x more output tokens for the same GDPx score
dylan522p · x · 2026-11-11
SemiAnalysis flags Claude Fable 5.2's poor token efficiency from persistence and subagent spawning: on GDPx materials-science tasks it scores 21.8% vs GPT-6 Astra's 21.6% but costs 89% more ($78.73 vs $41.72 per task) with 4.3x more output tokens; vs Gemini 4 Carbon ($10.43) the gap is 7.5x. The leaderboard also shows Claude Opus 5.5 at 16.5%, GPT-6 Luna at $0.89/task, and open models GLM 5.3 / DeepSeek 4.1 Flash below 3%.
Related event: SemiAnalysis GDPx Leaderboard: Claude Fable 5.2 Tops but at High Cost(3 posts)→
More from Venture
- SoftBank posts $7B quarterly loss after turning $2B Intel stake into $10.5B and hedging it — realsohamparekh · 2026-11-11
- Menlo Ventures: enterprise AI deployment spending hits ~$177B, 5.4x growth, agents dominate token use — ingliguori · 2026-11-11
- Inkbox AI raises $5.3M to give agents their own email, phone and WhatsApp — EXM7777 · 2026-11-11
- OpenAI acquires year-old travel startup Odessia; founder joins to work on travel — julianweisser · 2026-11-11
- Sahil Lavingia, who built the IRS iOS app in 3 months: build first, pitch later — every · 2026-11-11
- Analyst: As models converge, Anthropic and OpenAI will push deeper into verticals — RihardJarc · 2026-11-11