来源时间 本站刊期
[AINews] Fable 5.2 retakes World #1 LLM spot as $ANT IPO roadshow begins
Readers weighing Anthropic's IPO roadshow should note that Fable 5.2 reclaiming the top LLM spot lands on the same day, raising the cost-of-thinking question for buyers.
判断
Enterprise buyers and platform leads now decide whether Fable 5.2's 20% lower per-token price is worth its higher per-task bill: at $88.15 per GDPx task versus $67.16, they must pin the default Medium effort, budget for extra verification tokens, or keep Opus 5.5 for cost-sensitive work.
依据
Anthropic shipped Fable 5.2 and Mythos 5.2 on its IPO roadshow day, claiming a step above Opus 5.5 in almost all evaluated capabilities; the launch retook the #1 spot Gemini 4 Carbon had held for 10 days. List price fell 20%, from $10/$50 to $8/$40 per 1M input/output tokens, and Medium is the default effort in the Claude apps and Claude Code. SemiAnalysis's GDPx professional-task benchmark and Artificial Analysis both find the model more token-hungry: SemiAnalysis paid $88.15 per task versus $67.16 for 5.1, because where Opus 5.5 and Fable 5.1 tried about 20 test cases, 5.2 sometimes wrote hundreds or thousands and spun up subagents to check its work. That checking helped it take first place at 16.4%. Anthropic reports Fable 5.2 at High matched Opus 5.5 at Max on FrontierCode 2.0 for about half the cost per task. Subscription access is restricted: Ultra and Team Premium resets scale with tenure (1 under 6 months, 2 for 6-12, 3 above 12), usable until March 8.
- Fable 5.1 GDPx cost per task$67.16
- Fable 5.2 GDPx cost per task$88.15
- Input $/1M tokens, 5.1 to 5.2$10 to $8
- Output $/1M tokens, 5.1 to 5.2$50 to $40
没写清 The source gives no GDPx per-task cost at Medium or High effort, so it is unclear whether changing the default effort setting closes the $67.16-to-$88.15 gap.
下一步 Track SemiAnalysis's next GDPx rerun: if Fable 5.2 at Medium or High lands at or below $67.16 per task, the token-hunger penalty is an effort-setting artifact rather than a real price increase.