With OpenAI o1 & o3 Models Devs Can Set Reasoning Effort
This allows for more granular management of token use, inference speed level of explainability.
Reasoning models are important, as in entails of a process decomposing a complex tasks into smaller tasks via a process of internal reasoning.
This leads to more accurate responses, and in cases where the reasoning is shown, a significant level of inspectability and observability.
I think one of the contributing factors of DeepSeek Chat capturing the imagination of the masses was the their Chat UI streamed the model’s reasoning in real-time prior to issuing the final answer on output.
In the case of OpenAI, their reasoning models do not explicitly show their internal reasoning process to the user.
It needs to be stated that the user is charged for the tokens used for internal reasoning. Having access to the internal reasoning would help.
But, instead, OpenAI provides the final response based on their reasoning effort level (low, medium, or high).
In Short
Hidden Internal Reasoning: The model performs logical steps internally, but users only see the end result.
Higher reasoning_effort = More Explanation: At high effort, the model provides more detailed reasoning, but it still doesn’t reveal its exact thought process like a step-by-step debug log.
No Transparency into Model Internals: Unlike some AI systems that display intermediate reasoning steps (e.g., chain-of-thought prompting), OpenAI’s standard models don’t expose this.
You can however prompt the model to explain its reasoning (e.g., “Show your step-by-step thought process.”).
Reasoning Effort
OpenAI included 𝘳𝘦𝘢𝘴𝘰𝘯𝘪𝘯𝘨_𝘦𝘧𝘧𝘰𝘳𝘵 with the o3-mini model which offer devs more control over AI responses, reasoning length & cost.
It is a parameter that adjusts the depth of AI reasoning. You can now choose between:
Low — Faster responses, minimal reasoning, fewer tokens
Medium (Default) — Balanced depth and efficiency
High — Deeper reasoning, more detailed explanations
This feature gives businesses & developers more flexibility. Need quick answers for customer support? Go with 𝘭𝘰𝘸. Want detailed insights for complex problem-solving? Choose 𝘩𝘪𝘨𝘩.
This is a step toward more customisable AI interactions — helping users tailor responses based on their needs.
Below is a working Python notebook where the effort_levels variable can be set to high, medium or low.
pip install openai==0.28
import openai
import getpass
# Prompt user for OpenAI API key
api_key = getpass.getpass("Enter your OpenAI API key: ")
openai.api_key = api_key
# Define the question to test reasoning_effort
question = "Why is the sky blue?"
# Define reasoning effort levels: low, medium, or high.
effort_levels = ["high"]
# Function to get response and token usage
def get_response(reasoning_effort):
try:
response = openai.ChatCompletion.create(
model="o3-mini", # Updated model name
messages=[{"role": "user", "content": question}],
reasoning_effort=reasoning_effort # Apply the effort level
)
reply = response["choices"][0]["message"]["content"]
tokens_used = response["usage"]["total_tokens"]
return reply, tokens_used
except Exception as e:
return f"Error: {e}", None
# Generate and display responses for each effort level
for effort in effort_levels:
response, tokens = get_response(effort)
print(f"\n--- Reasoning Effort: {effort.upper()} ---")
print(response)
print(f"🔢 Tokens Used: {tokens}")And below the output where the reasoning effort is set to low, medium or high.
--- Reasoning Effort: LOW ---
The sky appears blue primarily due to a phenomenon known as Rayleigh scattering. When sunlight reaches Earth's atmosphere, it collides with molecules, small particles, and gases. Sunlight is made up of a spectrum of colors, each corresponding to different wavelengths. The shorter wavelengths (like blue and violet) are scattered much more strongly than the longer wavelengths (like red and yellow).
Even though violet light has an even shorter wavelength than blue and is scattered more, our eyes are more sensitive to blue light, and some of the violet light is absorbed by chemicals in the atmosphere. This combination results in the sky primarily taking on a blue hue.
🔢 Tokens Used: 148Notice how the tokens used increase.
--- Reasoning Effort: MEDIUM ---
The sky appears blue because of a phenomenon called Rayleigh scattering. When sunlight hits Earth's atmosphere, it interacts with air molecules and small particles. Sunlight is composed of a spectrum of colors, each corresponding to a different wavelength. Short-wavelength light (blue and violet) is scattered much more efficiently by these small particles than longer-wavelength light (red and orange).
Although violet light is scattered even more than blue, our eyes are more sensitive to blue light, and some of the violet light is absorbed by the upper atmosphere. As a result, the scattered light that reaches our eyes is predominantly blue, making the sky appear blue during the day.
🔢 Tokens Used: 471And reasoning effort set to high.
--- Reasoning Effort: HIGH ---
The sky appears blue mostly because of a phenomenon called Rayleigh scattering. When sunlight enters Earth's atmosphere, it collides with molecules and small particles. Sunlight is made up of a spectrum of colors, each corresponding to light waves of different lengths. Blue light waves are shorter and scatter much more than the longer red wavelengths. This scattering spreads blue light in all directions, making the sky seem predominantly blue to our eyes.
There are a couple of key points that explain this effect:
1. Sunlight contains all colors of the spectrum. When it passes through the atmosphere, it interacts with air molecules and tiny particles.
2. Rayleigh scattering is more effective at scattering light with shorter wavelengths (like blue and violet) than those with longer wavelengths (like red and orange). However, even though violet light is scattered even more than blue, our eyes are less sensitive to violet, and some of it is absorbed by the upper atmosphere.
3. The net result is that blue light becomes the dominant color we see when we look up at the sky.
This combination of factors—wavelength-dependent scattering and the sensitivity of our eyes—leads to the familiar blue color of the daytime sky.
🔢 Tokens Used: 898The graph below shows the tokens used between the tree settings.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
