Does a token buy you more or less now than it did a few months ago? We built a consumer price index (CPI) for AI coding output from Anthropic's Opus 4.6 model in SWE-chat, Feb 5–Apr 15, 2026. What we find looks like tokenflation:

The goods in our CPI basket are standard engineering outcomes like PRs, but we also included knowledge capture (any time work is stored somewhere durable), which shows the most positive trend and which resonates with us as heavy users of these tools.

Nominating “Tokenflation” for early-decision word-of-the-year for 2026.

I love the notion of 'token-level purchasing power'.

Still, it's really hard to create a stable 'basket of goods' when the frontier of model abilities is changing so quickly. Much like there is no automotive analog of a Tesla FSD from the 90s, a lot of what coding agents can do now could simply not have been done even a few months ago!

My vibes-based take is that I'm getting more value from these tools over time. Whether that's happening on a token-amortized basis is unclear, because more and more tokens are being spent on hidden reasoning.

Our CPI includes a hedonic adjustment based on the code survival rate. This goes from about 90% to 95% in our time period, so we incorporate a uniform 5% adjustment. This softens the tokenflation; code output seems to be improving!

To give you a concrete sense for what the above declines mean, here are the estimated outcomes for a 10K-token session:

Feb 2026: a third of a PR, 630 lines of code drafted, 8 files touched, 1 knowledge capture.

Apr 2026: a sixth of a PR, 91 lines of code drafted, 4 files touched, less than half a knowledge capture.

What is causing the decline in token-level purchasing power? One relevant thing we know is that the job of an Opus 4.6 token changed dramatically in this period, from purely writing code in February to coding, thinking, and explaining in mid-April:

We would love to expand our analysis to other models, providers, and time periods. We've created a Claude Code tool/plugin that you can use to analyze your own usage. The tool shares no data with us, but perhaps we could all pool our anonymized data to learn more about the true value of our tokens: https://github.com/bigspinai/plugins

We ourselves derive significant value from these tools and wonder whether measuring other relevant outcomes would change the CPI. That said, it seems clear that AI is becoming more expensive in ways that might be outpacing its current value.

I love the notion of 'token-level purchasing power'.

That said, it is really hard to create a stable 'basket of goods' when the frontier of model abilities is changing so quickly. Much like there is no automotive analog of a Tesla FSD from the 90s, a lot of what coding agents can do now could simply not have been done even a few months ago!

My vibes-based take is that I'm getting more value from these tools over time. Whether that's happening on a token-amortized basis is unclear, because more and more tokens are being spent on hidden reasoning.

Another link to the associated blog post: https://bigspin.ai/resources/the-decline-of-token-level-purchasing-power

@ChrisGPotts a token price index (TPI) is genius, thanks for doing this

@ChrisGPotts if you want another opinion on things to include in the basket of goods, happy to help

@ChrisGPotts A 90->95% survival rate should do more than increase PPP by 5%, the last 10% of PRs are 10x more valuable. You can't really do an inflation curve without actual price signals

@ChrisGPotts this is an amazing project btw

@ChrisGPotts i knew it

Tokenflation is the most honest thing to come out of AI research in months. You are paying more tokens for less output as model providers optimize for engagement over quality. The subscription fees go up, the CLAUDE.md files get longer, and the actual productivity gains remain a study in a deck somewhere.

@ChrisGPotts This is like updating from Windows 95 to Vista and complaining that everything takes longer while everyone else updated their hardware (model) and now has a 2x more capable system. You simply didn't update the hardware, the OS works.

@ChrisGPotts Tokenflation is the right frame. The bill only looks cheap if each token still buys useful engineering work.

@ChrisGPotts the framing that might matter more: what's the total cost to reach a fixed quality bar, not cost per line of output. if the first draft gets worse but the fix loop is faster, the CPI looks bad while the real outcome improves. time-to-acceptable-PR feels like the right unit