The Kimi K3 Moment

Kimi K3 时刻

I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart. Same tasks, same quality of output, and near identical token counts to get there. I expected an open model to be sloppier or to grind through more tokens on the way to the same answer, and neither turned out to be true.

在日常编码工作中，我一直将 Kimi K3 与 Claude 放在一起使用，实际上我根本无法区分它们。同样的任务、同样的输出质量，达到目标所需的 Token 数量也几乎完全一致。我原以为开源模型会更粗糙，或者在得出相同答案的过程中需要消耗更多的 Token，但事实证明这两点都不成立。

The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units. The subscription side is even more lopsided. Kimi’s paid plans start at $19 a month, and the $39 coding tier is far more generous than anything Claude sells anywhere near that price. Claude’s plans are metered tightly enough that a normal day of agent work can chew through the allowance before lunch.

两者的价格差距巨大。K3 的 API 价格为每百万输入 Token 3 美元，每百万输出 Token 15 美元。而 Claude 的顶级模型在相同单位下的价格分别是 10 美元和 50 美元。订阅方面则更加失衡：Kimi 的付费计划起价为每月 19 美元，其 39 美元的编程套餐比 Claude 同价位提供的服务要慷慨得多。Claude 的计划限制非常严格，以至于一天的正常智能体工作量在午饭前就可能耗尽额度。

Then there’s the fine print. Claude couldn’t sustain Fable access on the twenty dollar plan, so they turned it off, and the plan quietly falls back to Opus. When the headline model on your plan can be switched off because the economics don’t work, the plan was never really selling you the headline model. Kimi’s tiers don’t come with that asterisk.

此外还有细则问题。Claude 无法在 20 美元的计划中维持对 Fable 的访问权限，于是他们关闭了该功能，该计划悄然退回到 Opus 模型。当计划中的主打模型因为经济效益不佳而被随时关闭时，这意味着该计划从未真正向你出售过那个主打模型。而 Kimi 的套餐则没有这种“附加条件”。

Step back and the bigger story is what an unmitigated failure US AI policy has been. The administration held Fable back, and what finally shipped is a hindered version that refuses whole categories of work. Meanwhile a frontier quality model with none of those restrictions is a download away, released by a Chinese lab the US government has no ability to regulate. Whatever the theory behind gating American models was, it plainly wasn’t thought through, because the only people the gates constrain are American customers.

退一步看，更大的背景是美国人工智能政策的彻底失败。政府阻碍了 Fable 的发布，最终推出的版本却是一个受到限制、拒绝处理多种任务的版本。与此同时，一个没有这些限制、具备前沿质量的模型只需下载即可获得，而它是由美国政府无法监管的中国实验室发布的。无论限制美国模型的理论依据是什么，显然都没有经过深思熟虑，因为这些限制唯一约束的对象就是美国消费者。

Semgrep found GLM 5.2 beating Claude on their cyber benchmarks for exactly this reason. The restricted model declines the work and the open one just does it. And it isn’t only Kimi. GLM 5.2 came out under an MIT license, beats the latest Opus release on real work while not even claiming to be frontier, and costs a fraction of it.

Semgrep 发现 GLM 5.2 在其网络安全基准测试中击败了 Claude，原因正是如此：受限模型拒绝工作，而开源模型直接完成了任务。不仅是 Kimi，GLM 5.2 在 MIT 许可下发布，在实际工作中击败了最新的 Opus 版本，且并未自称是前沿模型，其成本仅为后者的一小部分。

OpenAI got pushed through the same government gauntlet with GPT-5.6 but came out the other side able to put their flagship on the twenty dollar plan. Whatever you think of OpenAI, they have runway here that Anthropic clearly doesn’t. I think I can see where this goes. The government will try to regulate AI and open source in particular, and it will run the playbook it ran for the auto industry. Decades of subsidies, bailouts, and protective tariffs produced American carmakers that sell trucks at home and barely register anywhere else in the world.

OpenAI 经历了同样的政府审查考验（GPT-5.6），但最终能够将其旗舰模型放入 20 美元的计划中。无论你如何看待 OpenAI，他们显然拥有 Anthropic 所不具备的缓冲空间。我预见到了事态的发展：政府将试图监管人工智能，特别是开源领域，并会重演汽车行业的剧本。几十年的补贴、救助和保护性关税，造就了只在国内销售卡车、在世界其他地方几乎毫无存在感的美国汽车制造商。

I expect the current administration to reach for the same tools here. Public private partnerships propping up domestic models that only get used inside the US and can’t compete internationally. That’s a sad future where America is the one country without access to the best models at the best prices, buying models deeply tied to the corrupt Trump administration that are neither the highest quality nor the cheapest. Until then, at least, I can’t come up with a reason to keep paying for Claude.

我预计现任政府会在这里使用同样的手段。通过公私合营来支撑那些只能在美国国内使用、无法在国际上竞争的本土模型。这是一个悲哀的未来：美国将成为唯一一个无法以最优价格获得最佳模型的国家，被迫购买那些与腐败的特朗普政府深度捆绑、既非质量最高也非价格最低的模型。至少在目前，我想不出任何理由继续为 Claude 付费。