Explaining the Surge in Codex Usage
jasondeanlee · x · 2026-07-14
This repost explains why a user saw their usage "explode" by focusing on Codex's billing and sub-agent behavior:
- Long context billing: Some claim that gpt-5.6 doubles in cost after exceeding 272k tokens. However, the cited content clarifies that subscriptions don't simply double the charge for longer contexts; the real cause lies elsewhere.
- Model trajectory length: The overall trajectory length (the sum of contexts across multiple compactions) isn't heavily impacted by reasoning effort, and output quality doesn't significantly differ after crossing the 272k context mark.
- Value of long context: A larger context offers three main benefits:
- Faster speeds, due to less frequent compaction;
- The ability to handle massive inputs;
- Potential cost savings when the system is well-tuned and cache hits are ideal.
- Compounding reasons for usage spikes: Another cited source notes that a combination of issues in Codex leads to abnormal usage/costs:
- gpt-5.6 becomes more expensive above 272k;
- Ultra sub-agents continue spawning Ultra, causing nested sub-agent consumption;
- Sol and Terra run on the v2 sub-agent layer, which is still in early stages and duplicates the full long context instead of starting a new history.
Overall, it details the implementation specifics of context, sub-agents, and billing in coding agents like Codex, and how these designs collectively amplify token consumption.
More from coding & agent
- Claude Code + Monid MCP builds a $12K/year B2B prospecting engine on live data — aryanXmahajan · 2026-10-26
- Karpathy shares CLAUDE.md rulebook to stop AI coding agents from piling up tech debt — mdancho84 · 2026-10-26
- Rethinking dependencies in the agent era: drop libs that save 20 lines of code — gethackteam · 2026-10-26
- RustGIS hits 2,000 GitHub stars in two weeks with agent-first architecture — giswqs · 2026-10-26
- Vercel Ships 12 New Products at Its 2026 Conference as It Races Rivals for the Agent Entry Point — FinanceYF5 · 2026-10-26
- Build an AI agent from scratch in under 50 lines of pure Python: ReAct loop tutorial — adnan_hashmi · 2026-10-26