Step-by-Step: Perplexity API Key

 Perplexity AI offers a specialized API platform focusing on search-grounded LLM responses (Sonar models). Like OpenAI, they use a tiered system where your limits increase as you spend more.

---

1. Step-by-Step: Generate Perplexity API Key

The API is managed through the Perplexity Console, which is separate from the standard chat interface.

1. Access the Console: Go to docs.perplexity.ai or directly to perplexity.ai/settings/api.

2. Add Payment Method: Unlike the chat "Pro" subscription, the API requires prepaid credits.

   • Navigate to the Billing tab.

   • Add a credit card and purchase a minimum of $5 in credits.

3. Generate Key: * Go to the API Keys section.

   • Click + Generate.

   • Copy the key immediately. It starts with pplx-. Perplexity will not show it again for security reasons.

4. Set Up Auto-Top-Up (Optional): To prevent your scripts from breaking when credits run out, enable "Automatic Top-up" to refresh your balance when it falls below $2.

---

2. API Usage Tiers

Your "Tier" is determined by your total cumulative spend on the platform.

Tier	Total Credits Purchased	Status	Monthly Spend Limit
Tier 0	$0	New/Trial	$5 (One-time)
Tier 1	$50+	Light Usage	$100
Tier 2	$250+	Regular Usage	$500
Tier 3	$500+	Heavy Usage	$1,000
Tier 4	$1,000+	Production	$5,000
Tier 5	$5,000+	Enterprise	Custom ($200k+)

---

3. Model-Wise API Usage Limits

Perplexity uses Requests Per Minute (RPM) as the primary throttle. Limits for Tier 1 are shown below; these scale significantly as you reach Tier 3 and above.

Model Category	Specific Models	Tier 1 RPM	Tier 3 RPM
Sonar (Standard)	sonar, sonar-pro	150	1,000
Reasoning	sonar-reasoning, sonar-reasoning-pro	150	1,000
Deep Research	sonar-deep-research	10	40
Async Search	POST /v1/async/sonar	10	40
Search (Raw)	POST /search	50 QPS*	50 QPS

*QPS = Queries Per Second. The Search API has a high-burst capacity of 50 requests instantly.

Important Implementation Notes

• Deep Research Limits: The sonar-deep-research model has much lower rate limits because it performs multiple autonomous steps (searching, browsing, and reasoning) per single request.

• Error Handling: If you receive a 429 Error, Perplexity recommends using the x-ratelimit-remaining headers in the API response to calculate exactly when your "bucket" will refill.

• Context Windows: While models like sonar-pro support large contexts, sending very large prompts frequently may trigger token-based throttling even if you stay under the RPM limit.