Skip to main content
Every AI strategy on Everstrike uses a specific model to evaluate markets and decide whether to trade. Your model choice has a direct impact on two things: how quickly each evaluation completes and how many credits each evaluation consumes. Models with deeper reasoning produce richer analysis but cost more and take longer. Faster models cost less and suit strategies that check often. Select the model that best fits your strategy’s evaluation frequency, complexity, and cost budget.

​
Available Models

Speeds listed are averages, not guarantees; a dash means the model has not been benchmarked yet. Actual evaluation time also depends on prompt length and the amount of data available to the model during each run. What that costs in practice: with GPT-6 Luna, a strategy that checks one market every hour costs about 2 cents a day, and one that checks every minute costs about $0.06 an hour (each check costs the model’s price once for the strategy plus once per market). The Preview tab shows the expected cost of any strategy before you submit it.

​
Evaluation Frequency Limit

Strategies with an evaluation frequency of 15 seconds or less can only use GPT-4.1-nano. This restriction ensures evaluations complete fast enough to keep up with the chosen frequency.

​
Choosing the Right Model

Different strategies call for different models. Use the guidance below as a starting point:

Most Strategies

GPT-6 Luna is the default: tuned for trading and inexpensive at $0.0005 per evaluation, typically answering in 2–5 seconds. Start here unless you have a specific reason to switch.

Lowest Cost

GPT-5-nano with minimal reasoning has the lowest cost per evaluation, at the price of a few seconds per check. It suits strategies that check infrequently.

More Analytical Power

GPT-5-mini with medium or high reasoning provides deeper market analysis. Use it for complex prompts where reasoning quality matters more than speed or cost.

High-Frequency (15s or Less)

GPT-4.1-nano is the only model available for strategies with an evaluation frequency of 15 seconds or less.
Restricting the data available to your strategy can reduce the cost per evaluation by up to 50% on most models (less on the reasoning models). Review the Supported Data page to remove data points your prompt doesn’t need, and add extra data sources only when your rules use them.