Topic
The metered per-token cost of running large language models, and the budgeting and measurement practices teams use to keep it predictable.
On October 13 Ramp gave outside businesses early access to Glass, the AI workspace it uses internally, with a spending meter on each project. The 30% bill cut co-CEO Eric Glyman cites comes from Router, a separate product.
Reality
beans-picker cut a median macOS agent task from $0.388 to $0.060 in its author's benchmark by replacing full UI-tree reads with short candidate lists. Whether that carries over depends on how much of your own agent's bill goes to reading the window.
Reality
An engineer writing on dev.to swapped the coordinating LLM in a multi-agent system for an XState machine and typed receipts, reporting a 70% token cut. Whether that transfers depends on what your coordinator cost.
Reality
A study of eight frontier models on SWE-bench Verified puts agentic coding at roughly 1,000 times the token cost of code chat, dominated by input, with the models' own pre-run estimates correlating no better than 0.39.
Reality
A post arguing the Model Context Protocol was built for 2024-era models pushed a 165-point Hacker News thread into a fight about token cost. The post reports no benchmark, so the measurement is left to your own agent.
Reality
Alibaba ran the reviewer internally for two years across tens of thousands of developers before releasing it. The write-up is specific about how files get selected and comments get positioned, and it stops short of publishing the benchmark scores.
Reality
Decodo's own runs on ten pages put raw HTML at 3 to 24 times the token count of the same pages as Markdown, with every model still answering correctly and the JSON-LD fields lost in the conversion.
Reality
Salesforce, FactSet, S&P Global and Moody's have all wired their data into Claude and ChatGPT, and only Moody's will name a charging method, which leaves per-seat contracts metering a unit that agents no longer respect.
Reality
Sergi Corruchaga reports his D-Engine harness produced the same diff as DeepSeek's official agent on 42 times fewer tokens, by letting the model propose SEARCH/REPLACE blocks and a local runtime apply them.
Reality
Consumption pricing made model access a variable cost, and McKinsey now watches it user by user, emailing the heavy spenders rather than cutting them off. EY reports a larger saving from a quieter method.
Reality
The 90 percent token saving is one engineer's own usage on his own repos, but the enforcement mechanism underneath it copies cleanly: a line-count threshold that returns a block, with a pipe-shaped hole in it.
Reality