API model string
gpt-5.5Context window
~1M tokens (922K input / 128K output)
Max output 128K tokens
- Knowledge cutoff
- December 2025
- Released
- 2026-04-24
- Modalities
- text, image in; text out
Last verified July 10, 2026
Rate limits by tier
OpenAI uses a five-tier spend-based system (Tier 1–5 plus Enterprise). All tiers are per-organization and shared across API keys within that org. Figures below describe GPT-5.5 — the live successor to the searched 'GPT-5' name, which was never a distinct production endpoint.
| Tier | Requirements | RPM | TPM | RPD | Concurrent | Notes |
|---|---|---|---|---|---|---|
| Free | No ongoing free production tier | not published | not published | not published | not published | GPT-5 family not available on Free tier per OpenAI docs. |
| Tier 1 | First payment (~$5 cumulative spend) | 500 (per third-party trackers — confirm in your console/dashboard) | 30,000 (per third-party trackers — confirm in your console/dashboard) | not published | not published | Early-2026 GPT-4o-class baseline; GPT-5.5 may have lower starting limits — confirm in dashboard. |
| Tier 2 | ~$50 cumulative + 7 days account age | higher than Tier 1 | higher than Tier 1 | not published | not published | Automatic advancement. |
| Tier 3 | ~$100 cumulative | not published | not published | not published | not published | Automatic advancement. |
| Tier 4 | ~$250 cumulative | not published | not published | not published | not published | Automatic advancement. |
| Tier 5 | ~$1,000 cumulative + 30 days | 10,000 (per third-party trackers — confirm in your console/dashboard) | 30,000,000 (per third-party trackers — confirm in your console/dashboard) | not published | not published | Manual limit-increase requests available at Tier 5 only; 3–10 business-day response. |
| Enterprise | Contact OpenAI sales | not published | not published | not published | not published | Funded startups may get elevated starting tiers via support. Provisioned throughput available. |
Swipe the table sideways to see every limit column.
- 1.Tier thresholds from third-party trackers (inference.net, aipricing.guru) — verify in OpenAI dashboard; official docs do not publish a stable RPM/TPM table.
- 2.Rolling 60-second and 24-hour windows (no fixed reset); OpenAI tracks RPM, RPD, TPM, TPD, and IPM.
- 3.Limits are per-model (some shared across a model family) and shared across all API keys in an org. Batch API has its own separate pool.
Limits verified against the OpenAI docs, July 10, 2026.
Token pricing
What you pay per million tokens (USD). Input and output are billed separately.
Input
$5.00
per 1M tokens
Output
$30.00
per 1M tokens
Cached input
$0.50
per 1M cached
Batch discount
−50%
on batch jobs
- Long-context surcharge: prompts exceeding 272K tokens bill the ENTIRE session at 2× input ($10.00/MTok) and 1.5× output ($45.00/MTok) — not just the portion above 272K.
- Priority tier: $12.50 input / $75.00 output per MTok.
- Batch pricing: $2.50 input / $15.00 output per MTok (50% off standard); Batch API pool is separate from live RPM quota.
- Flex pricing: same 50% discount as Batch but with best-effort real-time scheduling.
- Cache writes for GPT-5.6+ billed at 1.25× input rate.
- 10% data-residency uplift for regional processing on models released on or after March 5, 2026.
Side-project chatbot
$55
per month
Assumptions
5M input tokens / 1M output tokens per month
$5.00 × 5 + $30.00 × 1 = $25.00 + $30.00 = $55/month
Mid-scale app (70% cache hit)
$471
per month
Assumptions
60M input / 12M output; 42M tokens served from cache at $0.50/MTok
18M uncached × $5.00 = $90 + 42M cached × $0.50 = $21 + 12M out × $30.00 = $360; total $471/month
High-volume pipeline (Batch API)
$4,400
per month
Assumptions
400M input / 80M output per month at Batch rate
400M × $2.50 = $1,000 (Batch input) + 80M × $15.00 = $1,200 (Batch output) = wait — standard rate: 400M × $5.00 = $2,000 + 80M × $30.00 = $2,400; total $4,400/month (standard; Batch halves to ~$2,200)
Run your own numbers
Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.
Estimated gpt-5.5 spend
$300/mo
Input: $150
Output: $150
30M in × $5.00 + 5M out × $30.00 = $300
Same volume, priced across models
- Grok 4Cheapest$90
- Claude Opus 4.8$275
- gpt-5.5This model$300
Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to gpt-5.5 only. Estimates for comparison; real bills vary with request shape and long-context surcharges.
gpt-5.5 vs the alternatives
GPT-5.5 competes with Claude Opus 4.8 and Grok 4 (Grok 4.5) at the frontier tier.
| Aspect | gpt-5.5 | Claude Opus 4.8 | Grok 4 |
|---|---|---|---|
| Input price | $5.00/MTok | $5.00/MTok | $2.00/MTok |
| Output price | $30.00/MTok | $25.00/MTok | $6.00/MTok |
| Context window | ~1M (922K) | 1M | 500K |
| Cached input price | $0.50/MTok | $0.50/MTok | $0.50/MTok |
| Batch discount | 50% | 50% | excluded at launch |
| Long-context surcharge | 2×/1.5× above 272K | none | unpublished above 200K |
| Max output tokens | 128K | 128K (300K batch) | not published |
| Vision/multimodal | yes (text+image in) | yes | yes |
Swipe the table sideways to see every model.
Hitting a 429? The playbook
The exact errors you'll see
429 Too Many Requests{"error": {"message": "Rate limit reached for ...", "type": "requests", "code": "rate_limit_exceeded"}}HTTP header Retry-After (seconds to wait)x-ratelimit-limit-requestsx-ratelimit-remaining-requestsx-ratelimit-reset-requestsx-ratelimit-limit-tokensx-ratelimit-remaining-tokensx-ratelimit-reset-tokensWhy it happens & how to fix it
RPM exceeded — burst of requests hitting 500 RPM Tier 1 ceiling
Spread requests over time; add delay between calls; implement exponential backoff.
TPM exceeded — large prompts consuming 30K TPM Tier 1 budget
Set max_tokens conservatively (pre-admission token estimate counts toward TPM); split large documents; use Batch API for non-real-time workloads.
RPD daily cap hit at low tiers
Upgrade to paid tier; implement request queuing; shift workloads to off-peak hours.
Shared org API key pool exhausted by multiple services
Isolate high-traffic apps to separate org or API key; request Tier 5 limit increase.
Long-context requests (>272K tokens) consuming extra TPM via 2× billing multiplier
Keep prompts under 272K tokens; segment long documents or use chunking strategies.
Retry strategy
Honor the Retry-After header on every 429 response. Implement exponential backoff with jitter starting at 1 second, doubling each attempt up to 60 seconds. OpenAI uses rolling 60-second and 24-hour windows — not fixed clock resets — so you may only need to wait a few seconds. Use x-ratelimit-reset-requests and x-ratelimit-reset-tokens headers to predict the next available slot rather than sleeping blindly.
1import OpenAI, { RateLimitError } from "openai";23const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });45async function chatWithRetry(6 messages: OpenAI.Chat.ChatCompletionMessageParam[],7 maxRetries = 68): Promise<string> {9 let attempt = 0;10 while (attempt <= maxRetries) {11 try {12 const response = await client.chat.completions.create({13 model: "gpt-5.5",14 messages,15 max_tokens: 1024,16 });17 return response.choices[0].message.content ?? "";18 } catch (err) {19 if (err instanceof RateLimitError) {20 const retryAfter = Number(err.headers?.["retry-after"] ?? 0);21 const jitter = Math.random() * 1000;22 const backoff = retryAfter > 023 ? retryAfter * 100024 : Math.min(1000 * 2 ** attempt, 60_000) + jitter;25 console.warn(`429 — retrying in ${(backoff / 1000).toFixed(1)}s (attempt ${attempt + 1}/${maxRetries})`);26 await new Promise(r => setTimeout(r, backoff));27 attempt++;28 } else {29 throw err;30 }31 }32 }33 throw new Error("Max retries reached");34}How to raise your limits
The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.
Tier 1
AutomaticMake first payment (~$5 cumulative spend)
Unlocks: ~500 RPM / ~30K TPM (GPT-4o-class baseline; per third-party trackers — confirm in dashboard)
Tier 2
Automatic~$50 cumulative + 7 days account age
Unlocks: Higher RPM/TPM than Tier 1
Tier 3
Automatic~$100 cumulative spend
Unlocks: Higher limits
Tier 4
Automatic~$250 cumulative spend
Unlocks: Higher limits
Tier 5
Automatic advancement; manual limit-increase request thereafter (3–10 business-day response)~$1,000 cumulative + 30 days account age, then Settings → Limits → Request increase
Unlocks: ~10,000 RPM / ~30M TPM; manual ceiling raise beyond Tier 5 published defaults
Enterprise / Custom
NegotiatedContact OpenAI sales
Unlocks: Provisioned throughput, custom SLAs, elevated limits, funded-startup fast-track
Cut your token spend
Prompt caching
~90% cost reduction on repeated context ($0.50/MTok vs $5.00/MTok standard)Prefix-heavy workloads (system prompts, long documents) benefit automatically when matching prefixes are detected. Structure prompts so the static prefix comes first.
Batch API for non-real-time workloads
50% off all input and output ($2.50/$15.00/MTok)Batch API uses a separate pool that does not compete with live RPM quota. Ideal for classification, summarization, and data-enrichment pipelines.
Stay under 272K tokens per request
Avoid 2× input / 1.5× output long-context surcharge on the ENTIRE sessionSegment long documents; use chunking strategies; retrieve only relevant sections rather than feeding entire corpora.
Set max_tokens conservatively
Prevents TPM pre-admission quota wasteOpenAI's pre-admission token estimate (including max_tokens) counts toward TPM before the request executes. Overestimates waste quota headroom.
Use Flex pricing for latency-tolerant workloads
50% off (same as Batch) with best-effort real-time schedulingChoose Flex when you need real-time responses but can tolerate variable latency — same discount as Batch without asynchronous job management overhead.
Spread requests over rolling windows
Eliminates RPM spikes that trigger 429sOpenAI uses rolling 60-second windows. Stagger burst requests over 2–3 seconds to stay under RPM ceiling without adding artificial sleep.
Monitor x-ratelimit-remaining-* headers proactively
Prevents 429s before they occurBack off when x-ratelimit-remaining-requests or x-ratelimit-remaining-tokens approaches zero rather than waiting for a 429 error.
Org key isolation per product line
Prevents one service from exhausting another's quotaAssign separate API keys per product or traffic type so high-volume features cannot deplete limits for other workloads within the same organization.
Frequently asked questions
Does GPT-5 have its own API endpoint?
No. 'GPT-5' was never a distinct OpenAI API model string. As of July 2026, the flagship endpoint is gpt-5.5 (released April 24, 2026). Use model: "gpt-5.5" in your API calls.
What are the GPT-5 API rate limits?
GPT-5.5 (the live successor) starts at approximately 500 RPM and 30,000 TPM at Tier 1 (per third-party trackers — confirm in your OpenAI dashboard). Tier 5 reaches roughly 10,000 RPM and 30M TPM. These figures are from community trackers; official docs do not publish a stable per-tier table.
Is the GPT-5 API free?
No. GPT-5.5 is not available on a free tier. A first payment of roughly $5 is required to reach Tier 1 (paid access). Pricing is $5.00/MTok input and $30.00/MTok output at standard rates.
How do I increase my GPT-5 rate limits?
Limits advance automatically as your cumulative spend reaches $50 (Tier 2), $100 (Tier 3), $250 (Tier 4), and $1,000 + 30 days (Tier 5). At Tier 5, you can submit a manual limit-increase request via Settings → Limits in your OpenAI dashboard; expect a 3–10 business-day response.
What does the GPT-5 long-context surcharge mean?
Any request where the total session token count exceeds 272K tokens is billed at 2× the standard input rate ($10.00/MTok) and 1.5× the output rate ($45.00/MTok) for the ENTIRE session — not just the portion above 272K. Keep prompts under 272K to avoid this surcharge.
GPT-5 vs Claude Opus 4.8 pricing: which is cheaper?
Input pricing is identical at $5.00/MTok. On output, Claude Opus 4.8 is slightly cheaper at $25.00/MTok vs GPT-5.5's $30.00/MTok. Grok 4 (Grok 4.5) undercuts both at $2.00/$6.00/MTok. Claude Opus 4.8 also has no long-context surcharge, unlike GPT-5.5.
How can RapidDev help with GPT-5 API quota planning?
RapidDev's engineers help teams architect prompt caching, Batch API pipelines, and tier-progression strategies to minimize GPT-5.5 costs and eliminate 429 errors. Book a free scoping call at rapidevelopers.com/contact.
What is 'rate_limit_exceeded' in the OpenAI API?
It is OpenAI's error code returned in the 429 response body when you have hit your RPM, TPM, RPD, or TPD ceiling. The full error object looks like: {"error": {"message": "Rate limit reached for ...", "type": "requests", "code": "rate_limit_exceeded"}}. Check the x-ratelimit-reset-requests or x-ratelimit-reset-tokens header to see when quota resets.
Should I use the Batch API for GPT-5 to reduce costs?
Yes. The Batch API offers 50% off standard pricing ($2.50/$15.00/MTok) and uses a separate pool that does not compete with your live RPM quota. It is ideal for classification, summarization, and data enrichment pipelines where real-time responses are not required.
We build AI apps that don't hit rate limits
- Retry, backoff & caching built in
- Multi-provider fallback routing
- Fixed price, you own the code
30-min call. No commitment.