OpenAI Cut GPT-5.6 Sol's Thinking Budget 87% - Four Days After Launch
Sol's juice values dropped across every reasoning tier with no advance notice. Terra and Luna were left untouched.
OpenAI cut the GPT-5.6 Sol thinking budget four days after launch. Juice values - the internal name OpenAI uses for reasoning effort budgets in ChatGPT - dropped across every Sol tier with no advance notice. Sol (max) fell from 960 to 128. Every other tier followed, each roughly halved or worse.
| Model | Old Juice Value | New Juice Value |
|---|---|---|
| GPT-5.6 sol (max) | 960 | 128 |
| GPT-5.6 sol (xhigh) | 128 | 40 |
| GPT-5.6 sol (high) | 40 | 16 |
| GPT-5.6 sol (medium) | 16 | 8 |
| GPT-5.6 sol (low) | 8 | 4 |
After the cuts, a gap opened between Sol and the rest of the GPT-5.6 family. Terra (max) still runs at 960. Luna (max) sits at 768. Sol (max) now lands at 128 - below both cheaper models at their highest tiers, which is an unusual outcome for what OpenAI is calling a user-friendly change.
OpenAI Says Inference Optimizations Drove the Change, Not a Nerf
Thomas Sottiaux at OpenAI addressed the GPT-5.6 Sol thinking budget changes directly on X. "No nerfing, only good stuff!" he wrote. OpenAI ran inference optimizations and passed the savings back to subscribers - Sottiaux said that delivers about 10% more usage for ChatGPT Work and Codex users.
A second factor drove the revert. At launch, OpenAI expanded Sol's context window to 372k tokens, up from the 272k limit on GPT-5.5. That wider context caused more usage credits to be consumed than intended, so OpenAI rolled back the limit to 272k. The juice value reductions appear linked to that rollback.