AI / Core OpenAI Codex Application Fundamentals Interview Questions
What are the key differences between OpenAI's o-series reasoning models and the GPT series?
The o-series (o1, o3, o4) were OpenAI's dedicated reasoning models - designed to spend significant compute on hidden chain-of-thought before answering. The GPT-5.x series (mid-2025 onwards) has progressively integrated reasoning capabilities, creating a unified model line that adapts reasoning depth per task.
| Aspect | o-series (o1/o3/o4) | GPT-5.x (current) |
|---|---|---|
| Reasoning | Explicit separate thinking phase | Integrated; adaptive based on task complexity |
| Speed | Slower on simple tasks (always reasons) | Adaptive: fast on simple, deeper on complex |
| API | Mostly Chat Completions | Responses API recommended |
| Tool use in reasoning | Limited (some models) | Full interleaved thinking + tool use |
| Cost | Higher per-token (reasoning tokens billed) | Reasoning tokens included or separately billed |
| Status | Legacy; o4 models still active | Current recommended family |
# Legacy o-series (still supported but not recommended for new projects): response = client.chat.completions.create( model="o4-mini", messages=[{"role": "user", "content": "Find all bugs in this code: ..."}], reasoning_effort="medium", ) # Modern GPT-5.x with integrated reasoning (recommended): response = client.responses.create( model="gpt-5.5", input="Find all bugs in this code: ...", reasoning={"effort": "high"}, # same concept, unified API ) # gpt-5.3-codex: reasoning + coding fused in one model # No separate "thinking phase" - adapts dynamically response = client.responses.create( model="gpt-5.3-codex", input="Architect a microservices system for a real-time trading platform.", reasoning={"effort": "high"}, )
Key insight: the o-series were specialised branches of the model tree; the GPT-5.x series represents a unification where general intelligence, coding specialisation, and adaptive reasoning coexist in one model family. For most new projects, GPT-5.x supersedes the need to choose between a "smart reasoning model" and a "fast general model".
What is the key architectural difference between how o-series models reason vs how GPT-5.x models reason?
What does the 'reasoning_effort' parameter control in the o-series API?
More Related questions...
