Fable 5.2 发布：火柴人战斗测试生成慢动作打斗编排
Anthropic 放出了 Fable 5.2，同一套火柴人打斗测试它能编出慢动作镜头，比 Opus 5.5 快 1 分钟但花费 40.36 美元对 18.88 美元。做动画或动作编排生成的可以对比这两款再决定用哪个。
Daniel Zambrini 对 Anthropic 新发布的 Fable 5.2 进行了火柴人战斗测试，模型生成了包含慢动作镜头的完整打斗编排。成本方面，Fable 5.2 单次测试花费 40.36 美元，对照模型 Opus 5.5 为 18.88 美元，价格贵了一倍多。
测试在 Extra High 档位运行而非 Max 档位，依据是初步基准测试显示该档位效果最佳。尽管当时 Anthropic 服务器处于高负载状态，Fable 5.2 的完成时间仍比 Opus 5.5 快了约 1 分钟。
从结果看，Fable 5.2 在动作编排的复杂度上表现突出，慢动作等电影化镜头语言被直接整合进生成结果，代价是明显的成本差距。对预算敏感的用户可能仍倾向 Opus 5.5，而对编排质量要求高的场景会偏向 Fable 5.2。
2/ 同一个火柴人战斗测试，Fable 5.2生成了包含慢动作的打斗编排。 成本方面： Fable 5.2：40.36美元 Opus 5.5：18.88美元 Fable贵了一倍多，但在服务器高负载时，仍比Opus快了1分钟。 x.com/DanielZambrini… Daniel Zambrini @DanielZambrini FINALLY FABLE 5.2 IS HERE!!! And now we have the first Stickman fight test and as expected (or not) Anthropic outdid themselves. Output is simply amazing, with a superb choreography that includes even a slow motion. Cost ended up in 40.36$ vs 18.88$ from Opus. Time to complete surprised a bit as it was 1 minute faster than Opus even considering the ultra heave stress their servers are on right now Important to phrase that I tested it in Extra High instead of Max, as the preliminary benchmarks pointed that this was the best achieving effort.
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Fable 5.2, Opus 5.5, Anthropic, 视频生成.
Source attribution: curated from AI Will (@FinanceYF5). Cross-source corroboration: 11 independent sources have covered this development (including Claude, berryxia, 宝玉), confirming industry-wide signal strength.
