AI / Core OpenAI Codex Application Fundamentals Interview Questions
What is the OpenAI Codex and API pricing model and how do you estimate costs?
OpenAI uses a pay-per-token pricing model for API access. For Codex CLI and App, costs are consumed from your ChatGPT or API credits balance. Understanding the cost structure helps in designing cost-efficient applications.
| Model | Input price | Output price | Use case |
|---|---|---|---|
| gpt-5.5 | ~$5/1M tokens | ~$20/1M tokens | API general use |
| gpt-5.4-mini | ~$0.60/1M tokens | ~$2.40/1M tokens | Lightweight tasks, subagents |
| codex-mini-latest | $1.50/1M tokens | $6/1M tokens | CLI-optimised Q&A |
| gpt-5.3-codex-spark | Metered (high-throughput) | Metered | Pro plan only, 1000+ tps |
| Batch API (any model) | ~50% of standard | ~50% of standard | Bulk non-real-time workloads |
| text-embedding-3-small | ~$0.02/1M tokens | N/A | Embeddings for RAG |
# Cost estimation example: # Scenario: Code review agent processes 100 PRs/day # Average PR: 2000 input tokens + 500 output tokens = 2500 tokens daily_prs = 100 avg_input_tokens = 2000 avg_output_tokens = 500 # Using gpt-5.5 at standard pricing: input_cost_per_million = 5.00 output_cost_per_million = 20.00 daily_input_cost = (daily_prs * avg_input_tokens / 1_000_000) * input_cost_per_million daily_output_cost = (daily_prs * avg_output_tokens / 1_000_000) * output_cost_per_million daily_total = daily_input_cost + daily_output_cost monthly_total = daily_total * 30 print(f"Daily cost: ${daily_total:.2f}") print(f"Monthly cost: ${monthly_total:.2f}") # With Batch API (50% discount): batch_monthly = monthly_total * 0.5 print(f"Monthly cost (Batch API): ${batch_monthly:.2f}") # With prompt caching (40% cache hit rate, 75% discount on cached): # Effective input cost reduction ~30% # + Batch API = significant overall saving
Cost reduction hierarchy: choose the right model for the task (biggest lever) > use prompt caching > use Batch API for bulk > use smaller models for sub-tasks > use streaming to reduce perceived latency without changing cost.
What is the approximate cost saving when using the OpenAI Batch API vs synchronous API calls?
Which cost reduction strategy typically has the biggest impact on overall API spend?
More Related questions...
