Portfolio 14 · quota envelope

LLM API rate-limit and concurrency planner

Find the minimum of request, input-token, output-token, concurrency, and daily-budget capacity, then show how retries and explicit quota windows change the result.

Decision this answers: Which documented or account-specific limit binds steady traffic, and will the actual burst fit its rolling or fixed window?

Scenario

Decision summary

Safe steady capacity

160.75exact-derived

Binding constraint: daily budget.

[1][2][3]

Daily request plan

231481exact-derived

Steady capacity multiplied by 1,440 minutes.

[1]

Retry quota multiplier

1.08model estimate

Expected attempts counted against all applicable quotas.

[1]

Example burst headroom

12.5%exact-derived

A 3,500-request window against the 4,000 RPM example.

[1]

Direct-labeled result profile

Safe steady capacity Daily request plan Retry quota multiplier Example burst headroom

The tables below are the accessible source of truth; bar lengths never carry identity alone.

Quota window ledger

Quota window ledger
Candidate	Request quota	Token quota	Window	Retry treatment	Guarantee	Evidence
Anthropic	Tier/account-specific	Input/output categories	Documented rolling behavior	Counts by failure point	Quota compliance only	published list[1]
OpenAI	Account/project-specific	Model-specific	Response headers/dashboard	Retries consume capacity	Quota compliance only	published list[2]
Gemini	Tier/project-specific	Model-specific	Documented tier windows	Retries consume capacity	Quota compliance only	published list[3]

Official surface versus this site

Anthropic rate limits

Use the official surface to confirm native prices, counting rules, limits, and product eligibility.

This offline workbench

Provider docs define quota categories and windows. This page reconciles them with measured latency, request shape, retry behavior, and a daily budget without claiming provider availability.

Sources and methodology

Every result-affecting reference is visible here without JavaScript and is retained in the JSON export.

[1] Anthropic rate limits

Authority: Anthropic
Native identifier: rate limit tiers and token buckets
Unit: RPM and token throughput
Evidence: published list
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Quota categories and rolling behavior remain provider-specific; displayed defaults are examples.

Open the authoritative source

[2] OpenAI rate limits guide

Authority: OpenAI
Native identifier: rate limit headers
Unit: requests and tokens
Evidence: published list
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Account-specific values are local overrides and retries consume capacity.

Open the authoritative source

[3] Gemini API rate limits

Authority: Google
Native identifier: Gemini API tiers
Unit: requests and tokens
Evidence: published list
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Tier and project scopes remain explicit; examples are not availability guarantees.

Open the authoritative source

Limits and non-claims

Quota-compliant capacity is not a latency, throughput, or availability guarantee.
Account-specific defaults are examples and user overrides stay local.
Backoff scheduling, jitter, and the selected failure point must match the client implementation.