Processing 14.2B tokens with GPT-5.6 Sol without hitting the Pro plan limit
nodo48 · reddit · 2026-09-01
The author reports processing 1.4212 billion input tokens over 28 days using the Codex Pro plan ($200 tier, utilizing only 5.6 Sol extra high). The system employs a fail-closed architecture, certifying tasks only upon objective proof, and achieved a 98.293% cache hit rate. The author is surprised that this high volume has not yet hit the usage limits of the Pro plan and asks for similar experiences or explanations.
More from coding & agent
- Anthropic Releases List of 17 Free Official Claude Courses — ZabihullahAtal · 2026-09-01
- 7 Agentic Patterns: Picking the wrong one breaks your workflow — mdancho84 · 2026-09-01
- Training a PPO Agent with Grok to Play a Self-Built Game — tetsuoai · 2026-09-01
- Two people manage 13M creators using a custom internal Agent OS — lxfater · 2026-09-01
- Voice-agent latency: What metrics matter after fixing streaming? — asgillette · 2026-09-01
- OpenAI Swarm失控揭示审计层缺失风险 — Master-Sprinkles-848 · 2026-09-01