GPT-5.6 Sol uses twice the tokens of GPT-5.5

GPT-5.6 Sol xhigh used 2.25x more tokens per session than GPT-5.5 xhigh in Codex workflows. Comparing the same 14-day period for each from deduplicated local logs, the number of sessions only increased 2.9%, from 1,667 to 1,715.

• GPT-5.5 xhigh: 12.17B tokens total, 7.30M tokens per session
• GPT-5.6 Sol xhigh: 28.22B tokens total, 16.45M tokens per session

Overall usage also increased 2.32x, from 12.17B tokens to 28.22B tokens. With a similar number of sessions consuming more than double the usage quota, the practical value of the token-based quota dropped to less than half.

The author explained that previously, even with extreme workloads, three subscriptions would last about a week, but after GPT-5.6 Sol became the top-tier xhigh model, a subscription was exhausted in a single day of moderate-level work. GPT-5.5 and GPT-5.6 Sol have the same fresh-input, cached-input, and output token pricing, but GPT-5.6 Sol also added a cache-write fee that GPT-5.5 didn't have.