OpenAI's Internal Juice Budget Reportedly Shifts

johnseach · x · 2026-07-13

This post explains the so-called juice values: hidden internal budgets that dictate how much "thinking compute" a model can spend per request.

• Higher juice allows the model to perform more reasoning steps, usually performing better on complex problems, but it's slower and more expensive.
• Lower juice means faster, cheaper responses, but weaker performance on multi-step reasoning tasks.
• The community often "probes" these values using specific diagnostic prompts, especially within Codex-related environments.
• The post also claims that the maximum juice for GPT-5.6 Sol was recently slashed from around 960 to 128 as part of efficiency optimizations, while Terra and Luna retained higher budgets.

Original post →

More from Models

• GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
• 105 hidden bugs, 2 repos: DeepSeek V4.1 Flash fixes 24 at $1.80 vs Opus 5's 27 at $51.33 — ChartsJournalX · 2026-09-11
• awesome-llm-leaderboards: an open-source directory of LLM leaderboards, pricing tables, comparison tools — Last_Establishment_1 · 2026-09-11
• Anthropic claims it works to keep eval environments unidentifiable to models — MaxKannen · 2026-09-11
• Nex N2.5 Pro released on Hugging Face with 407GB of weights — jinnyjuice · 2026-09-11
• RoMa v2 image matching model unveiled in the usual black poster — ducha_aiki · 2026-09-11

Models channel →