CODEXIS AI

Pro reasoning mode for OpenAI models

The Pro reasoning mode for OpenAI models in CODEXIS AI: what it does, how it is billed and how many credits it uses in production.

You switch the reasoning mode in Model settings under Reasoning mode, which has the values Standard and Pro. The option is available for the models GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna and GPT-6 Astra, and the default value is Standard.

Like the other model settings, you choose the mode before starting the conversation; it cannot be changed in a chat that is already in progress. This page explains what Pro does, how it is billed and how many credits it uses in everyday work.

What Pro mode does

CODEXIS sends the selected mode to OpenAI as the reasoning mode of the request. According to OpenAI's documentation, in Pro mode the model performs more work on your request than in standard mode, which increases token usage and cost. It is intended for difficult tasks where you can accept a longer wait and higher consumption.

Reasoning mode and Reasoning effort are independent of each other according to OpenAI. The mode selects standard or Pro execution, while the effort controls how much reasoning the model applies within that mode. You can therefore combine Pro with any effort level.

An answer in Pro mode takes noticeably longer. CODEXIS waits up to 30 minutes for a single model response.

How Pro is billed

Pro has no separate price list. Every token is billed at the same credit rates as in standard mode, according to the selected model and the Tier setting, see Models and credit rates.

It costs more because the model produces far more reasoning tokens, usually a longer answer, and in practice needs more steps for a single question. Reasoning tokens are billed at the output rate, which is the most expensive one.

In the Price estimate in the model settings, Pro counts as thirty times the price of standard mode. The actual consumption follows the size of the input and the length of the answer, see the examples below.

How much Pro uses in practice

Medians from ten days of production, chat requests of the main model:

ModelCredits per request, StandardCredits per request, ProRatioReasoning tokens, StandardReasoning tokens, Pro
GPT-5.6 Terra5905,3009x2902,950
GPT-5.6 Sol2,60026,60010x4704,160
GPT-6 Astra4,80037,6008x3801,900

For a whole question including all model steps, GPT-6 Astra shows a median of 18,200 credits in standard mode and 189,400 credits in Pro mode, roughly ten times as much. Questions asked in Pro mode also carry about 3.5 times more input tokens, a median of 929,000 against 267,000. The examples in the next section therefore belong among the smaller Pro questions.

These ratios describe how people use Pro, not the effect of the mode alone. They turn it on for longer documents and longer conversations, so the input grows as well. On GPT-5.6 Terra, at the same input size, a request in Pro mode costs about 1.5 to 2.6 times a request in standard mode and produces three to ten times more reasoning tokens.

In the observed period, Pro accounted for 1.4% of user questions and 35% of all credits consumed.

Production example: GPT-6 Astra

All three examples are real production questions on the GPT-6 Astra model with the Tier setting Standard and standard context. One request is one call of the model within a single user question, for example after the model has searched CODEXIS. The credit totals match the amount actually deducted from the account.

Rates for GPT-6 Astra in the Standard tier with standard context per 1,000,000 tokens:

TierInputInput cachedOutput
Standard250,000 credits25,000 credits1,250,000 credits

Output includes reasoning tokens.

Pro, reasoning effort high

High is the default value of Reasoning effort.

RequestInputOf which cachedOutputOf which reasoningCredits
155,73140,7321,2771,0076,364
262,17843,2923,3901,56310,195
365,40348,1561,3928477,256
Total183,312132,1806,0593,41723,815

Standard, reasoning effort high

A question of comparable size, three requests and roughly 183,000 input tokens.

RequestInputOf which cachedOutputOf which reasoningCredits
156,63055,7583511192,100
262,58856,5265575163,997
363,37162,5425132952,461
Total182,589174,8261,4219308,558

On the output side, meaning the answer including reasoning, the difference is 6,059 against 1,421 tokens, or 7,574 against 1,776 credits. The rest of the difference lies on the input side, where the standard question had a larger share of its input served from cache, 174,826 of 182,589 tokens against 132,180 of 183,312. The full 2.8x difference in price therefore cannot be attributed to Pro alone.

Pro, reasoning effort maximum

RequestInputOf which cachedOutputOf which reasoningCredits
126,12320,1241,7221,4874,231
232,47424,3482,7572,5846,086
337,57128,0607,8756,26012,923
Total96,16872,53212,35410,33123,241

At a rate of 1.25 credits per token, the reasoning tokens alone cost 12,914 credits, that is 56% of the price of the whole question, on an input of under 100,000 tokens.

When to turn Pro on

Turn Pro on for demanding analysis where the quality of the answer matters more than speed and price. For ordinary work, leave the mode at Standard and first try raising Reasoning effort, which is usually enough. Watch your consumption in the credit count shown under each answer.

The other model settings are described in The model and its customization, and billing in Credits and pricing.