If you're registering for OpenAI's Codex or ChatGPT and need a phone number to complete SMS verification, the common advice on forums is "just pick whatever country is cheapest." The real-world fulfillment data says otherwise: delivery rates for the same platform vary wildly depending on where the number originates.
This post shares one small operator's records — 31 paid buyers and 100+ activations across 26 countries between August and September 2026 — plus how to interpret them without marketing hype. Sample sizes are small; treat this as directional engineering data, not gospel.
Quick answers from fulfillment data
- US and Japan numbers had the highest paid-buyer delivery rates in this dataset: 4 out of 5 buyers received their code after switching to a dedicated US/Japan-focused pool (n=5, small cohort caveat applies). During the earlier broad multi-country period, 15 of 26 buyers completed verification (57.7%).
-
Per-activation success clustered around the top countries:
- United States: 26.7% (n=15 attempts, average wait time 164s)
- Japan: 25.0% (n=12 attempts, average wait time 487s)
- Colombia: 25.0% (n=16 attempts)
- South Africa: 18.2% (n=11 attempts)
- Chile: 16.7% (n=12 attempts)
- Portugal performed worst in this dataset: only 5.4% across 37 activations. Twenty other countries hovered between 0% and 7%.
- Cheap numbers correlate with recycled segments: Numbers sold for $0.01–$0.05 on wholesale aggregators failed at the highest rates.
Why "cheapest" is the wrong filter
Public SMS marketplaces sort by price by default, so the top of every list is dominated by numbers from segments that have been queried hundreds of times. OpenAI's fraud and risk models flag those carrier ranges aggressively.
The result is exactly what the per-activation records show: a long tail of near-zero delivery rates in bargain tiers, and a narrow set of countries where fresh telecommunications prefixes remain unflagged.
Practical takeaways from the data:
- Prioritize clean prefixes over bargain pricing: US and Japan routes consistently showed cleaner reputation scores, even though wholesale unit cost is higher ($0.30–$0.66).
- Stop after two consecutive failures: If a country route fails twice, switch regions immediately. Spamming attempts against an already-throttled prefix triggers temporary rate limits on your registration IP.
- Account for delayed delivery: Japan's average delivery latency was 487 seconds (~8 minutes). Users who cancel after 60 seconds often discard numbers that would have successfully received the code.
- Bind authenticator app (TOTP) and recovery email immediately: A temporary number is strictly for initial SMS verification, not persistent account recovery. Without immediate TOTP binding, any future session challenge could lock the account.
The honest caveats
- Sample size: 100+ total activations, and the "switched to US/Japan pool" subset rests on 5 buyers. This is empirical operator fulfillment telemetry, not an academic controlled study.
- Risk rule volatility: OpenAI updates its risk heuristics frequently. What delivered reliably last month can degrade as wholesale segments burn out.
- Code delivery != long-term account health: Completing phone verification does not exempt an account from IP reputation scoring, fingerprint consistency, or payment gateway checks.
Further Reading & Telemetry
- For the full interactive table breaking down delivery times and activation stats across all 26 countries, see the research breakdown on NumHarbor.
- For Chinese-language troubleshooting and step-by-step registration failure fixes, refer to the Codex SMS Guide.
Disclosure: The author operates Codex SMS (shanma.store), where these operational fulfillment logs were recorded. All data is reported as observed without promotional exaggeration.
Top comments (0)