Credit-Based vs Per-Call Tiered Search Pricing for AI Agents
Credit-based pricing works better for agents because query volume is unpredictable. Per-call tiers punish agents that spike occasionally but average low volume. Understanding the cost model difference is the difference between predictable spend and surprise overages.
How Per-Call Tiered Pricing Works
SerpAPI charges: 250 free, then $25/1k, $75/5k, $150/15k, $275/30k per month. You buy a tier, use up to that amount, and roll to the next tier if you exceed it.
The problem for agents: tier boundaries are harsh. An agent that uses 1,001 calls in a month pays the 5k tier price ($75) for those 1,001 calls — an effective rate of $0.075/call rather than $0.025/call. The next 3,999 calls are essentially free (already paid for), but if you never use them, you have overpaid for the tier.
For human-driven workflows with predictable volume, tiers work fine because you choose a tier matching your expected usage. For agents, usage depends on how many tasks users trigger, how complex those tasks are, and whether the agent decides to run additional searches mid-task. None of these are predictable.
How Credit-Based Pricing Works
Scavio charges $0.005/credit. Each search call costs 1 credit. If you make 1,001 calls, you pay $5.005. If you make 4,000 calls, you pay $20. There is no tier to buy into and no unused capacity to waste.
Tavily: $0.008/credit PAYG, or $30/4k credits ($0.0075/credit). Exa: $7/1k standard searches = $0.007/call.
For agents, the key property is linear cost scaling. Double the queries, double the cost. No step-function surprises.
Actual Cost Comparison at Different Volume Points
Assuming monthly query volumes:
| Monthly queries | SerpAPI cost | Scavio cost | Tavily PAYG |
|---|---|---|---|
| 200 (free tier) | $0 | $1.00 | $1.60 |
| 1,000 | $25 | $5.00 | $8.00 |
| 2,500 | $25 | $12.50 | $20.00 |
| 5,000 | $75 | $25.00 | $40.00 |
| 10,000 | $75 (if on 15k plan) or $150 | $50.00 | $80.00 |
| 30,000 | $275 | $150.00 | $240.00 |
At every volume, credit-based pricing is cheaper. The tradeoff: SerpAPI's free tier (250/month) is generous for evaluation. Once you exceed the free tier, credit-based wins economically.
The Agent Variance Problem
Consider an agent running 10 tasks/day that each uses an average of 5 searches. Normal volume: ~1,500 searches/month. You buy the $25/1k SerpAPI tier. But during a busy week, 50 tasks run and each uses 10 searches due to complex queries: 500 searches in a week. Monthly total: 2,000+. You now need the $75 tier.
With Scavio at $0.005/credit, the same variance costs $7.50 in a normal month and $10+ in the busy week. No tier upgrade, no surprise invoice.
Serper's Free Tier Exception
Serper gives 2,500 free queries as a one-time allocation. If you are building and testing and your total evaluation queries stay under 2,500, Serper is genuinely free. For production use beyond that, it shifts to $50/yr for 50k queries ($0.001/query annually, or $50/mo for 500k). Serper's annual plan is the cheapest option at high volume if your query type (basic Google search) is supported.
Credit Pool Mechanics Matter
Some providers sell credits in prepaid blocks that expire. Others (Scavio, DataForSEO) use a deposit model where credits roll over. For agents that might run heavily one month and lightly the next, rollover credits are materially better than expiring blocks.
Check the expiration policy before buying a large credit block. Expiring credits create the same over-spend incentive as tier pricing — you feel pressure to use them before they disappear.
Recommendation by Agent Type
Research agents (variable query volume, research tasks of different complexity): credit-based (Scavio, Tavily PAYG)
Monitoring agents (fixed queries on a schedule, predictable volume): annual plans with rollover, or Serper annual for basic Google search
Evaluation and prototyping (under 1,000 queries): Tavily free tier (1,000/mo) or SerpAPI free tier (250/mo)
High volume production (100k+ queries/month): DataForSEO ($0.002/call) or Serper annual plan