OpenAI replaced five paid API usage tiers with Build, Launch and Grow on October 6, 2026. Organizations advance automatically as total credit purchases reach each tier’s threshold. For teams taking an application into production, the change makes qualification easier to read—but launch planning still needs separate checks for spending and throughput.
What qualifies an organization
The current usage-tier table lists these thresholds and monthly ceilings in US dollars:
Tier | Total credit purchases required | Approved monthly usage ceiling |
|---|---|---|
Build | $5 | $500 |
Launch | $100 | $5,000 |
Grow | $500 | $200,000 |
Free remains separately listed with geographic eligibility and a $100 monthly ceiling; that is not a promise of $100 in free credits. Paid-tier thresholds are cumulative purchases, not monthly subscription prices. Grow’s $200,000 ceiling is not included credit or a recommended budget.
Choose spending controls separately. Under OpenAI’s spend-limit rules, alerts notify without stopping requests; enforced organization or project caps interrupt affected traffic with HTTP 429. An organization cap covers all its projects. Enforcement can lag slightly, so spend can exceed the configured amount. A team qualifying for Grow can still choose a much lower cap, with alerts before that cutoff.
Token limits can bind before request limits
OpenAI publishes these Standard limits for GPT-6 Astra and GPT-6 Luna; its general table gives Sol and Terra the same bands as Astra. RPM means requests per minute; TPM means tokens per minute.
Tier | Astra / Sol / Terra RPM | Astra / Sol / Terra TPM | Luna RPM | Luna TPM |
|---|---|---|---|---|
Build | 5,000 | 1,000,000 | 5,000 | 2,000,000 |
Launch | 10,000 | 4,000,000 | 10,000 | 10,000,000 |
Grow | 15,000 | 40,000,000 | 30,000 | 180,000,000 |
Illustrative calculation, not a benchmark: assume one Standard Astra workload, no competing traffic, and exactly 10,000 tokens charged against its TPM allowance per request. Using the published Astra limits, the arithmetic ceiling is min(RPM, TPM / 10,000): 100 requests per minute on Build, 400 on Launch and 4,000 on Grow. Thus Build’s 5,000 RPM figure would not let this workload send 5,000 such requests per minute. These are rate-limit-accounted tokens, not an assertion that billing counts identically. Actual throughput can be lower; this is not a latency or capacity guarantee.
Do not apply that table to Ultrafast. Astra’s default Ultrafast allowances are 500,000 TPM on Build, 1 million on Launch and 5 million on Grow, with separate pricing.
Three checks before raising production traffic
Confirm effective limits. Open Settings → Organization → Limits. Check project overrides, shared model pools and long-context limits, not just the tier label. Public documentation does not establish every account’s migration timing or overrides.
Ramp gradually. A
429 slow_downcan occur below RPM/TPM ceilings. FollowRetry-Afterwhen present, reduce traffic, then increase it gradually. OpenAI’s ramp guidance is an operational constraint, not a promised speedup from changing tiers.Handle billing failures separately. Inspect
error.code: configured caps, the approved monthly usage ceiling and an exhausted credit balance need different remedies. Decide who can change a cap, request more approved usage or add credits; repeated retries alone will not restore access.
For teams managing project configuration in code, the related RohitAI Terraform guide explains rate-limit records and spend alerts, including where those controls stop.
Methodology: AI-assisted reporting and analysis based on OpenAI’s published documentation, rechecked October 6, 2026. No customer dashboard, account migration or API performance was tested.