DISCLOSURE: If you buy through affiliate links, I may earn a small commission. (disclosures)
I've been experimenting with GPT 5.6 + OpenCode for the past couple weeks, running GPT 5.6 Sol through OpenAI's $100/month ChatGPT Pro (5x) plan.
Before this I only used the Claude Max plans for personal work. I wanted a direct usage comparison between the subscriptions but mostly found "it depends" and "apples to oranges" responses, so here's the post I wish existed - how much GPT 5.6 usage I actually got from the $100/month sub.
How Much GPT 5.6 Usage I Got
In 3 days I used 75% of my weekly quota. That came out to an API-equivalent price of ~$413 based on ccusage:
- Logged tokens - 407M
- Cached input - ~90%
Projecting that 75% to the full quota comes out to about 543M tokens or $550 in API-equivalent usage per week:
- Per week - ~$550
- Per month - ~$2,200
Calculation Caveats
This is an early read from one setup for one week so this anecdata comes with a lot of asterisks.
- Small sample - this is 3 days of anecdata on one account for one usage pattern so it's not enough for a strong statsig read on generalized API-equivalent costs.
- Cache rate - ~90% of my input was cached. A lower cache hit rate could make the same number of tokens much more expensive at API prices.
- Model + harness - I mostly used OpenCode with GPT 5.6 Sol, various subagents (Terra, Luna, and Claude Opus), and my own custom skills. Different setups may get different ROI.
- Local readout -
ccusageonly sees logs available on this device so it may not capture everything perfectly. I typically use multiple devices though I was traveling at the time so most usage was on this device.
Next
My early data suggests OpenAI's GPT subscriptions are heavily subsidized against API prices, similar to Claude's. The gap likely overstates the labs' actual losses since some estimates peg inference margins at 60% and subscriptions also work as marketing for eventual enterprise API adoption (ppl like me get the subsidized sub, build workflows / get attached to the harness+model, and bring them back to their org).
If you want to see my setup, I snapshot my ai-dotfiles each month for HAMINIONS Members.
Read next: