Anthropic 发布 Claude Sonnet 4.5

2025 年 9 月 29 日，Anthropic 发布了 Claude Sonnet 4.5。这场发布之所以在开发者圈子里炸开，不是因为「又变强了」这种常规叙事，而是因为它挑战了一个默认前提：旗舰就该是又大又贵的。Sonnet 4.5 用中等尺寸，在代码和 Agent 场景里打出了接近甚至超过大旗舰的成绩，价格和延迟却低一大截。

2025 年的旗舰竞赛，主线是「更长、更难的任务」，各家都在用超大参数和大算力往前冲。但与此同时，开发者的抱怨也在积累：旗舰确实强，可也实在太贵，Agent 跑一次任务烧掉的 token 让人心疼。市场需要的是「既能扛住 Agent 场景、又算得过账」的模型——Sonnet 4.5 精准地补上了这个位置。

从评测和社区反馈看，它做到了。在编程、自主任务这类「干活」场景，Sonnet 4.5 与 Opus 4 级别的模型互有胜负，部分场景反超，成本却只有零头。开发者社区很快把它捧成了「写代码首选」，高频、大批量的 Agent 应用尤其受益。

Sonnet 4.5 的成功把「性价比」推到旗舰竞争的中心。它证明在能力逼近之后，「单位成本能换来多少能力」才是开发者真正关心的。这种逻辑反过来倒逼大旗舰在价格与效率上让步，也让「小而精」成为各家的新军备竞赛方向。

回看 Sonnet 4.5，它的历史价值在于验证了一个判断：尺寸不是一切。当工程效率、成本控制和场景适配做到极致，中等尺寸的模型也能赢得头部之战。这场「小而精」的逆袭，改写的不仅是 Anthropic 的产品线，更是整个行业对旗舰的定义。

On September 29, 2025 Anthropic released Claude Sonnet 4.5. The launch exploded across developer circles not for the routine "it got stronger" narrative but for challenging a default assumption: that flagships must be big and expensive. Sonnet 4.5, at mid size, delivered near or above large-flagship results in coding and agent scenarios at far lower cost and latency.

The 2025 flagship race ran on "longer, harder tasks", with everyone charging ahead on massive parameters and compute. Meanwhile developer complaints accumulated: flagships were strong but painfully expensive, and an agent run burned through tokens painfully. The market needed a model that could both carry agent workloads and make financial sense—and Sonnet 4.5 filled exactly that slot.

Judging by evals and community feedback, it did. In "getting work done" scenarios like programming and autonomous tasks, Sonnet 4.5 traded wins with Opus-4-class models and beat them in some, at a fraction of the cost. Developers quickly crowned it the "coding first choice", and high-frequency, high-volume agent apps benefited most.

Sonnet 4.5's success pushed cost-performance to the center of flagship competition. It proved that once capabilities converge, what developers really care about is "how much ability per dollar". That logic forced big flagships to concede on price and efficiency, and made "small and sharp" the new arms race direction.

Looking back, Sonnet 4.5's historical value lies in validating one judgment: size isn't everything. When engineering efficiency, cost control, and scenario fit are done to the extreme, a mid-size model can win the flagship war. This "small and sharp" upset rewrote not just Anthropic's lineup but the industry's definition of what a flagship is.