Cloudflare AI Gateway — What the Free Tier Includes
Cloudflare AI Gateway is an observability and control layer placed between your application and AI providers (OpenAI, Anthropic, Gemini, Workers AI, Replicate, etc.). Pointing provider API calls at the gateway — a one-line code change — gives you analytics, caching, rate limiting, and fallback.
It is available on all plans, and the core features are free. The only meaningful free-tier boundary is log storage volume, so small to medium AI apps can effectively run at no cost.
Features included for free
| Feature | Description |
|---|---|
| Analytics | Dashboard visibility into request counts, token usage, and cost |
| Logging | Stored logs of requests and errors (limits below) |
| Caching | Serve responses from Cloudflare's cache instead of re-calling the provider — cuts both latency and provider billing |
| Rate limiting | Cap incoming request volume |
| Retry / model fallback (dynamic routing) | Automatic retries and failover to another model or provider on error |
| DLP | Data loss prevention scanning — free on all plans (2 predefined profiles without a Zero Trust subscription) |
| Logpush | Export logs to external destinations (R2, S3, etc.). The feature is free; exported volume is usage-billed |
Workers AI can be a routing destination behind the gateway. Workers AI itself offers 10,000 Neurons/day free, so the two free tiers combined let you build an AI app PoC at near-zero cost.
Free-tier limits
Gateways and caching
| Item | Limit |
|---|---|
| Gateways | 10 per account |
| Gateway name | 64 characters |
| Cacheable request | up to 25 MB per request |
| Cache TTL | up to 1 month |
| Custom metadata | 5 entries per request |
Rate limiting
- Unified Billing throttle: 200 requests per 60 seconds per gateway, returning
429when exceeded. - Applies only to Cloudflare-managed credentials, not BYOK (your own provider API keys).
Log storage — the real free-tier boundary
Logs are where AI Gateway consumes resources, and the rules depend on when your account created its first gateway (rules changed September 24, 2026).
| Regime | Applies to | Free plan limits |
|---|---|---|
| Legacy Logs | Accounts that created their first gateway before 2026/09/24 | 100,000 logs total across all gateways (10 MB per log, 500 logs/sec per gateway) |
| Workers Logs | Accounts creating their first gateway on/after 2026/09/24 | 200,000 events/day, 3-day retention |
Upgrading to Workers Paid ($5/month) raises this to 10,000,000 logs per gateway (Legacy Logs) or 20 million events/month ($0.60 per additional million) under Workers Logs. For long-term retention, export via Logpush.
Workers Logs pricing is scheduled to move under Cloudflare Observability pricing on December 1, 2026. If you are creating a new gateway, check the current pricing page.
Items billed separately
Three things fall outside the free tier:
| Item | Billing |
|---|---|
| Guardrails | Billed as Workers AI token-based inference (depends on prompt + response length) |
| Logpush | Billed by exported/transformed log volume |
| Unified Billing | 5% fee on credit purchases (e.g. $105 for $100 of credits). Provider inference rates pass through without markup |
Summary
- Core features (analytics, caching, rate limiting, fallback, DLP) are free on all plans.
- The free tier is bounded only by log storage: 100,000 logs for legacy accounts, or Workers Logs (200k events/day, 3-day retention) for gateways created on/after 2026/09/24.
- Billing risks are limited to four items: log volume overage, Guardrails, Logpush export volume, and the 5% Unified Billing credit fee.
- Provider inference costs still apply — the gateway adds no markup of its own.