Revolut 表示，在不到一年的时间里，通过将自托管模型作为默认选择，而非主要依赖 Claude 等商业服务，该公司已将 AI 编码推理成本削减了约 90%。英国金融科技公司 Revolut 的运营与基础设施负责人 David Abitbol 上月在纽约举行的 AI Engineer 大会上发言时表示，在向约 2000 名工程师推出 AI 编码工具后，编码时间提升了 60%，前置时间缩短了 70%。
Revolut says that in less than a year it's managed to slash its AI coding inference costs by roughly 90 percent by making self-hosted models its default instead of relying primarily on commercial services like Claude. Speaking at the AI Engineer conference in New York last month, David Abitbol, ops and infrastructure lead at the British fintech, described how it saw a 60 percent improvement in coding time and a 70 percent improvement in lead time following its rollout of AI coding tools to roughly 2,000 engineers.
Revolut 最初采用了 GLM-5.1，随后迁移至 GLM-5.3，目前使用 56 块 Nvidia B200 GPU 进行服务。在转向 GLM 之前，Revolut 曾尝试过 AI，于 2025 年初部署了 Gemini，2026 年初部署了 Claude Code。“我们对 AI 的成本持高度怀疑态度。我们看到工程师每月花费 1000 美元、1 万美元，却没有多少投资回报，因此我们在推出时设定了严格的预算限制，”Abitbol 说。因此，工程师只能使用有限数量的 token，公司便开始研究自托管自己的模型。
Revolut initially adopted GLM-5.1 before moving to GLM-5.3, which it now serves using 56 Nvidia B200 GPUs. Prior to its switch to GLM, Revolut had been experimenting with AI, deploying Gemini in early 2025 and Claude Code in early 2026. "We were highly skeptical about the cost of AI. We were seeing engineers spend $1,000 a month, $10,000 a month with not so much ROI, and so when we launched it we launched it with strict budget limits," Abitbol said. Because of this, engineers only had access to a limited number of tokens, so the company began investigating self-hosting iyd own models.
自托管 GLM 5.1 的成本为每百万输入/输出 token 0.30 美元，约为使用 OpenAI Codex 或 Anthropic Claude Code 成本的十分之一。结果证明，当 token 变得便宜时，你就不再担心使用多少 token，也不担心用在什么地方。
At $0.30 per million input/output tokens, self hosting GLM 5.1 cost about a tenth of using OpenAI's Codex or Anthropic's Claude Code. And, as it turned out, when tokens are cheap, you stop worrying about how many they use or what they use them for.
虽然 Revolut 确实看到了编码速度的提升，但 Abitbol 指出，最大的瓶颈其实并非代码本身。他说，最大的收益来自于让 AI 模型处理调试、Jira 和 Confluence 工单等事务。“当你拥有无限 token 时，你就可以自由地做这些事，”他说。“你不再需要为季度末或最困难的任务节省 token。”
While Revolut did see improvements to coding speed, Abitbol noted the biggest bottleneck wasn't actually the code. Instead, he says the biggest gains came from letting the AI models deal with things like debugging and Jira and Confluence tickets. "When you have unlimited tokens, you're free to do that," he said. "You no longer have to save your tokens for the end of the quarter or the most difficult task."
GLM 5.1 由中国开放权重模型开发商 Z.ai 开发，并非一个小模型。
Developed by Chinese open weights model dev Z.ai, GLM 5.1 isn't exactly a small model.
总参数量约为 7540 亿（每生成一个 token 激活 400 亿参数），该模型仅存储权重就需要约 380 GB 的 GPU 显存，每个用户维护上下文还需额外 6-26 GB，具体取决于精度。为大规模部署 GLM-5.3，Revolut 部署了 56 块 Nvidia B200 GPU，GPU 数量相当于七套八卡 DGX B200 系统。但正如 Abitbol 所指出的，最大的性能提升并非来自增加 GPU，而是调整模型超参数、改用 NVFP4 量化、实现预填充与解码解耦，以及调优上下文长度。
At roughly 754 billion total parameters (40 billion activated for each token generated), the model requires roughly 380 GB of GPU memory just to hold the weights and another 6-26 GB per user to maintain context depending on precision. To serve GLM-5.3 at scale, Revolut deployed 56 Nvidia B200 GPUs, equivalent in GPU count to seven eight-GPU DGX B200 systems. But as Abitbol noted, the greatest performance gains didn't come from adding GPUs, but rather tuning the model's hyperparameters along with switching to NVFP4 quantization, implementing disaggregated prefill and decode, and tuning context lengths.
一个关键结论是，支持一百万 token 的上下文窗口，并不意味着如此大的窗口实用或高效。“长上下文是个骗局。你真的得留心这一点，”Abitbol 说。“即使你能服务一百万 token，25.6 万才是你真正需要的。” 经过这些优化，Abitbol 表示公司实现了在约 25.6 万 token 上下文窗口下、首 token 生成时间低于 3 秒，生成速度约为每秒 200 token，这两项指标均为其软件工程师的“最佳平衡点”。
One key takeaway was that supporting a one-million-token context window doesn't necessarily make such a large window useful or efficient. "Long context is a scam. You really need to be mindful of that," Abitbol said. "Even though you can serve one million tokens, 256 [thousand] is really what you need." With these optimizations, Abitbol says that the company was able to achieve sub 3-second time to first token on context windows of around 256,000 tokens, and generation speeds of around 200 tokens a second, both of which he notes are the sweet spot for the company's software engineers.
“你要追求的是首 token 时间低于 3 秒。这是实证。一旦超过 5 秒，我的 Slack 收件箱就会炸锅，”Abitbol 说。“吞吐量其实没那么重要。就交互而言，我们提供每秒 200 token。大家都能接受。没必要更高。人们的工作方式变了。我们不再盯着思维链 token 流。人们并行运行任务，任务完成再查看结果。” 他补充说，如果担心吞吐量，下午 5 点后工程师们能达到接近每秒 800 token。
"What you want to aim for is a sub-3 second time to first token. This is empirical evidence. As soon as you exceed 5 seconds, my Slack inbox starts to blow up," Abitbol said. "In terms of throughput, it doesn't really matter as much. We serve, in terms of interactivity, 200 tokens per second. People are fine [with that]. You don't need to go higher. The way people work has changed. We no longer look at the stream of thinking tokens.
不过值得注意的是，GLM 并未完全取代 GPT 或 Claude。该中国模型仅支持文本，因此不太适合 UI 等工作。
People run tasks in parallel and check in when the task is done." And if throughput is a concern, he notes engineers who after 5 pm are getting closer to 800 tokens a second. However, it's worth noting that GLM hasn't replaced GPT or Claude entirely.
Revolut 采用了模型路由器，将视觉查询转发给 Gemini 或 GPT‑Luna 等模型，同时让代码生成仍在 GLM 上运行。Abitbol 说：“开源现在是 Revolut 的首要任务，我们的经济模型也很有利。计费是固定的，月底不会出现账单冲击，前沿技术只是例外。”他补充道，虽然中国的开源权重模型在隐私和安全方面可能存在担忧，但主要问题是通过离开中国的 API 访问它们。若在本地或区域云资源上运行，则根本不必担心。 ®
The Chinese model is text only, so it isn't ideal for things like UI work. So instead, Revolut used a model router to direct vision queries to models like Gemini or GPT Luna while keeping code generation running on GLM. "Open source is now a priority at Revolut, and our economics are favorable. The billing is flat, we don't have any bill shock at the end of the month, and the frontier is the exception," Abitbol said. Having said that, Abitbol acknowledges that Chinese open weights models may be a concern for privacy and security reasons, but emphasizes that the major concern there is accessing them via APIs running out of China. When run on-prem or on regional cloud resources, there's really nothing to worry about. ®