Coinbase CEO wants engineers to continue using TokenMax while keeping AI costs down.
In a Friday post on X, CEO Brian Armstrong outlined five ways crypto exchanges can keep AI costs low.
The first of his five strategies was to choose a better default LLM (the model most engineers use by default when sending prompts). He said Coinbase is experimenting with a Chinese-made LLM as the default, which is significantly cheaper than models from leading U.S. AI labs like Anthropic and OpenAI.
“We are experimenting with defaulting to open-weight models like GLM 5.2 and Kimi 2.7 via the LLM Gateway, while also encouraging engineers to choose the model appropriate for the task,” Armstrong wrote.
GLM 5.2 and Kimi 2.7 are models developed by Chinese AI research institute Z.ai and Moonshot AI, respectively.
His second strategy, which he talked about in early June, is to route prompts to the most appropriate model based on difficulty.
“For example, a frontier model may be necessary for planning, but undesirable for execution because it may be overkill,” he writes. “Ultimately, humans shouldn’t choose models; AI can automate this task.”
The third tip was to use better caching, a technique to reduce inference costs.
Fourth, avoid wasting context, which means starting a new session when switching between tasks.
And his final strategy is to improve visibility into AI spending across the company. This means that all of his engineers can use as many tokens as they want, but can see their usage. Coinbase hopes that employees will see a “bigger impact” by spending more on AI.
At the end of his post, Armstrong included a chart tracking the company’s token usage and AI spending over time, but did not specify an exact timeline. This graph shows that while token usage has recently reached one of the highest levels in the company’s history, spending on AI has fallen significantly to nearly half of its peak.
“The goal is not to curb usage; it is to build infrastructure that makes rapid growth sustainable,” he wrote.
Armstrong’s post comes less than two months after Coinbase laid off 14% of its staff, in part as AI changes the way people work.
“Over the past year, we’ve seen engineers leverage AI to ship work in days that previously took teams weeks,” Armstrong said in a May post, adding, “The pace at which small, focused teams can accomplish it has changed dramatically.”
His strategy to reduce AI costs is consistent with the industry. The industry is moving away from the trend of short-term token maxing and favors imposing usage caps on employees to curb rampant token consumption.