Why Your AI Is Running on Low Power
OpenAI recently rolled out its powerful GPT-5.6 model family, including Luna, Terra, and Sol. These advanced models come equipped with a crucial internal parameter called reasoning_effort, which directly controls their computational depth and the "thinking" they apply to complex tasks. Think of it as a power setting, allowing the AI to engage in everything from quick, surface-level analyses to deep, multi-step problem-solving. This setting determines how thoroughly your AI processes information before delivering an output.
Here’s the critical oversight: the most potent "Max" reasoning setting, designed to unlock the full analytical prowess of these models, is not enabled by default. This applies universally to Luna, Terra, and Sol within the Codex platform's configuration. Without manual intervention, you are unknowingly running your sophisticated AI agents on a throttled setting, missing out on their true capabilities from the moment you start using them.
This default configuration significantly throttles your AI's performance, leading to suboptimal and often unsatisfying results. Even when you select a flagship model like Sol, specifically built for intricate coding challenges and extensive agent tasks, it operates below its true potential. You pay for the highest-tier processing power but receive outputs compromised by limited "thinking effort," effectively wasting valuable tokens and computational budget. For Luna, now 80% off, activating "Max" becomes an even smarter move to maximize its value.
The Luna Max Power Play
OpenAI recently slashed Luna's price by a staggering 80%, transforming this small language model (SLM) into a formidable cost-performance leader. This dramatic reduction makes Luna an incredibly attractive option for a vast array of tasks, especially those requiring high volume and cost sensitivity.
Even with Luna's new affordability, you are likely missing out on its full potential. By default, the Max reasoning effort for Luna, Terra, and Sol remains disabled in Codex. This means your AI agent isn't leveraging the deepest computational depth available for its price point.
Unlocking Luna's maximum thinking effort is straightforward. Navigate to Codex > Settings > Configuration. Under 'Available reasoning efforts', you will see options for Luna, Terra, and Sol. Go ahead and toggle the 'Max' setting to ON; it’s off by default.
Activating 'Max' immediately empowers Luna to engage its maximum thinking effort, all at that unprecedentedly low, 80% discounted price. This simple toggle transforms Luna's value proposition, delivering superior analytical capabilities without burning your budget. Avoid 'Ultra mode' for now; it currently consumes tokens inefficiently without tangible benefit.
Avoid This Token-Burning 'Upgrade'
While enabling Luna's Max reasoning effort is a smart move for cost savings, there's another setting in Codex that's currently a hidden trap: ultra mode. This feature, which aims to provide deeper, more exhaustive reasoning, is basically broken right now. Activating it will quickly burn through your tokens and inflate your OpenAI bill.
Documented issues reveal that ultra mode frequently gets stuck in infinite review loops. Imagine your AI endlessly re-evaluating the same task, unable to finalize an output. This cycle not only wastes compute cycles but also spawns numerous inefficient sub-agents, each consuming tokens without contributing to a meaningful resolution.
This isn't a true upgrade; it's a flawed mechanism that actively sabotages efficiency. Instead of providing superior results, ultra mode only leads to high costs and poor performance, directly undermining the cost-effectiveness we discussed with Luna's 80% price cut.
To keep your operations lean and your AI bill under control, you must turn this ultra mode setting OFF in your Codex configuration. It's a critical step in optimizing your AI pipeline for the GPT-5.6 family, especially when leveraging powerful tools like Codex in ChatGPT | AI Coding Agents for Software Engineering - OpenAI.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Your New Optimal AI Strategy
Your AI models, Luna, Terra, and Sol, now operate at peak efficiency with a precise configuration. Enable the Max reasoning effort for all three in Codex settings; this unlocks their full computational depth, moving beyond their default low-power state.
Crucially, disable ultra mode entirely. This setting is currently broken, burning your valuable tokens without delivering useful results. It's a costly misstep many users unknowingly make by leaving it active.
With these baseline adjustments, selecting the right model becomes straightforward. For speed and cost-sensitive workloads, leverage Luna Max. Luna's massive 80% price reduction makes it an exceptional value for high-volume tasks and AI evaluations.
Terra Max offers reliable, balanced performance for your everyday tasks. For the most complex agentic workflows, Sol Max provides the ultimate power and depth needed for large builds and long-running operations.
Blindly trusting default settings is a recipe for wasted resources in this fast-moving AI landscape. Actively configuring your models now ensures optimal performance and prevents unnecessary token expenditure, saving you money and maximizing your AI's potential.
Frequently Asked Questions
What are OpenAI's Luna, Terra, and Sol models?
They are part of the new GPT-5.6 family. Luna is a fast, low-cost model for high-volume tasks, Terra is a balanced model for general use, and Sol is the flagship model for complex reasoning and coding.
Why should I enable 'Max' reasoning in Codex?
The 'Max' setting provides the highest quality output from the models. It's disabled by default, but enabling it—especially for the 80% discounted Luna—offers a significant boost in performance for a very low cost.
What's wrong with 'Ultra mode' in Codex?
'Ultra mode' is currently reported to be inefficient and broken. It causes excessive token consumption, can get AI agents stuck in loops, and often fails to produce better results, making it a costly setting to use.
How do I change reasoning settings in Codex?
In the Codex platform, navigate to Settings, then Configuration. Under the 'Available reasoning efforts' section, you can toggle the 'Max' setting on for the models you use.