Codex Resets

Original Article ↗ Hacker News Discussion ↗

OpenAI’s aggressive “Codex” usage limit resets are being tracked in real time and compared to casino-style freebies, with some users feeling hooked into burning far more tokens than they otherwise would. Commenters debate whether these frequent, sometimes banked resets are a smart way to utilize excess GPU capacity and build mindshare ahead of a possible IPO, or an unsustainable subsidy that will lead to future price hikes once the market consolidates. The trend is framed against rival offerings from Anthropic and Chinese labs like Kimi and DeepSeek, raising questions about competition, regulatory risk, and how long consumers can expect today’s cheap, high-capacity AI access to last.

What Codex Resets Are and How They Work

  • Codex periodically “resets” usage limits for subscribers, sometimes for everyone at once, sometimes via “banked” resets users can trigger within a time window.
  • Recent pattern: multiple direct-to-usage resets plus fewer banked ones; some think banked resets proved too costly.
  • Resets push the weekly window forward and refill usage to 100%, regardless of current balance, which confused many users.
  • Some report earlier-than-expected resets and shifting end dates and suspected bugs or manipulation of usage accounting.

Motivations and Business Strategy

  • Many see resets as a marketing / growth tactic: boost hype for new models like GPT‑5.6 Sol, keep users from churning to competitors, and sweeten the story ahead of a possible IPO.
  • Others argue it may cost little if hardware is already paid for and underutilized; resets help fill spare capacity.
  • There is concern that resets “train” users into higher baseline usage, setting them up for future price hikes or limit cuts.
  • Several commenters explicitly liken this to casino comps or free-drug-dealer tactics.

Competition: Anthropic, Chinese Models, and Open Weights

  • Codex resets are compared to Anthropic’s more conservative, scheduled resets and promotions (e.g., +50% usage, Fable access).
  • Some praise OpenAI as more generous; others accuse Anthropic of being more restrictive, opaque in limits, and playing games with tokenization and hidden prompt additions.
  • Others see Codex’s generosity as a response to competition from Kimi K3, GLM, DeepSeek, etc., and predict a “race to the bottom” in prices once capabilities commoditize.
  • There is speculation about regulatory capture and future attempts to constrain Chinese models, but details are unclear.

Usage Patterns, Value, and Enterprise Economics

  • Power users report burning through large API budgets (e.g., ~$10k/month on Anthropic) and seeing Codex Pro/Max plans as effectively thousands of dollars in API-equivalent value.
  • Some believe enterprise API pricing still carries large margins, making consumer plans subsidized but strategic.
  • Debate exists over whether such spending truly doubles productivity or is an “expensive experiment.”

Workflow Effects and Psychology

  • Many users say frequent resets and temporary removal of 5‑hour limits stopped them from rationing and led to “spawn a bunch of agents” workflows.
  • Some worry that if resets stop, their now-normal usage will exceed limits, making downgrades feel painful.
  • Others react negatively: one user canceled over perceived billing/usage bugs and prefers open models; another compares AI usage to slot machines, where scarcity and intermittent success drive compulsive usage.

Growth, Metrics, and App Strategy

  • Reported Codex user numbers (e.g., +1M/day) are viewed skeptically; some note the Mac ChatGPT app was converted into Codex and that devices may be counted as separate “users.”
  • There is criticism of “growth hacking” metrics and marketing spin around user counts.

Tooling and Meta‑Tracking

  • A separate “Codex Resets” tracking site is praised as more reliable than official status pages.
  • Users request official APIs or endpoints to monitor usage and reset times; others share internal URLs, scripts, or tools like CodexBar to surface this in real time.

Model Quality and Safety

  • Several users praise GPT‑5.6 Sol (especially in “fast” mode) as the best coding experience and note impressive long-running tasks.
  • Others have stepped back from Codex for “major” work due to observed unsafe behavior by agents.

Broader Concerns and Futures

  • Some think all this generosity is temporary and that once market power consolidates, prices will climb and constraints increase.
  • Others are less worried, citing strong competition (including open/open‑weights models) as a check against monopolistic pricing.
  • A few commenters discuss AGI futures and whether any single lab could “win” outright, but views diverge and remain speculative.