Included usage pricing appears to have changed significantly for Kimi K3 Max and GPT-5.6 Sol Max

Support Help
1

I noticed something odd in the Included Usage page and compared several billing windows. I’ll attach screenshots.

Cursor’s own models look basically consistent when you normalize usage by tokens consumed per 1% of allowance:

  • Composer 2.5 Fast: ~40–41M tokens per 1%
  • Grok 4.6 XHigh: ~49–51M per 1%
  • Grok 4.6 High Fast: ~24M per 1%

So those appear stable.

But the same calculation gives a very different result for some “Other Models”.

For Kimi K3 Max:

  • Sep 5 window: 664.2M tokens = 68.1% → 9.75M tokens per 1%
  • Sep 9 window: 724.4M = 82.3% → 8.80M per 1%
  • Sep 15 window: 221.3M = 50.7% → 4.36M per 1%

That means Kimi appears to be consuming included usage at roughly 2x the previous rate now.

For GPT-5.6 Sol Max:

  • Earlier: 9.5M tokens = 1.7% → 5.59M per 1%
  • Current: 5.6M tokens = 2.6% → 2.15M per 1%

That implies roughly a 2.6x increase in effective included-usage cost.

The fact that Cursor models remain almost perfectly stable makes this look less like a display/rounding issue and more like a pricing or multiplier change for specific external models, apparently sometime around mid-September.

Was there a pricing/multiplier change for Kimi K3 Max and GPT-5.6 Sol Max that wasn’t announced, or am I misunderstanding how Included Usage is calculated?

Screenshots attached.

1
11093×471 14.7 KB

2
21095×515 16.1 KB

3
31098×750 22.7 KB

6

Another fresh data point from my account: Kimi K3 Max used 4.0M tokens and already consumed 3.8% of the “Other Models” pool — only about 1.05M tokens per 1%.

Cursor says Other Models are charged at the model’s API rate. Kimi K3’s official pricing is:

Cache-hit input: $0.30 / 1M
Cache-miss input: $3 / 1M
Output: $15 / 1M

So this 3.8% consumption for just 4M tokens looks unusually high unless a very large portion was output/reasoning. Screenshot attached.
image