Tavily vs Exa vs Perplexity Sonar API: Real AI Web Search Cost per 100,000 Queries in 2026

The short answer: Tavily and Exa both price simple search retrieval in the same narrow band, about $0.007–$0.008 per query, and Perplexity's flat Search API undercuts both at $0.005. The multiplier that actually separates these products is not the base search; it is deep content extraction and, especially, the research/deep-search endpoint. Tavily's research tier costs 4 to 250 credits per call, a 62.5x range within one product, which at 100,000 calls a month is the difference between $3,200 and $200,000. Perplexity's Sonar API, which returns a synthesized answer instead of raw results, adds a per-request fee on top of token charges, and the fee itself varies 2.4x by context size. None of these figures include a separate downstream LLM call to generate a final answer from raw search results, except where the product (Sonar) does that synthesis itself.

What each vendor bills

Tavily (OFFICIAL, Tavily API-credits documentation). Pay-as-you-go at $0.008 per credit. Basic search: 1 credit. Advanced search (deeper crawling): 2 credits. Extract, basic: 1 credit per 5 successful URL extractions. Extract, advanced: 2 credits per 5 extractions. Map: 1–2 credits per 10 pages. Research (mini model): 4–110 credits per request. Research (pro model): 15–250 credits per request. Subscription plans lower the effective per-credit rate at volume: Project ($30/mo, 4,000 credits, $0.0075/credit), Bootstrap ($100/mo, 15,000 credits, $0.0067), Startup ($220/mo, 38,000 credits, $0.0058), Growth ($500/mo, 100,000 credits, $0.005). A free Researcher tier gives 1,000 credits a month with no card required. New accounts get $20 in free credits (about 2,800 standard searches).

Exa (OFFICIAL, Exa pricing page). Fixed per-endpoint rates, unlike Tavily's variable credit weighting: $7 per 1,000 search requests ($0.007 each) and $5 per 1,000 answer requests ($0.005 each). New accounts receive $20 in starting credits (about 2,800 searches) plus $10 of free credits every month with no card required; requests simply fail closed when the balance runs out. Enterprise volume discounts exist but were not itemized (QUOTE-ONLY / UNKNOWN).

Perplexity Sonar API (REPORTED). A hybrid meter: $1 per million input tokens and $1 per million output tokens, plus a per-request fee of $5 to $12 per 1,000 requests depending on context size ($0.005–$0.012 per request). A separate flat Search API (no synthesis, no token charge) was reported at $5.00 per 1,000 requests ($0.005 each), the cheapest simple-search option in this comparison. Sonar itself returns a pre-synthesized answer with citations rather than raw results, so its price includes work that Tavily and Exa leave to a downstream LLM call.

The three query shapes

All ILLUSTRATIVE, using the vendors' own published units. A: simple search result retrieval — Tavily basic search (1 credit), Exa search ($0.007), Perplexity flat Search API ($0.005). B: search plus full-page content — Tavily advanced search (2 credits, includes deeper crawling), Exa search (content is included in the same $0.007 call), Perplexity Sonar at the low end of its context-based fee plus a modest token charge (500 input, 150 output tokens, ILLUSTRATIVE). C: research or deep-search style query — Tavily's research endpoint at both ends of its published range, Perplexity Sonar at the high end of its fee with triple the token usage (a longer, multi-source synthesis, ILLUSTRATIVE).

Cost per 100,000 queries

Formula: cost = queries × price per query, using pay-as-you-go rates throughout (subscription tiers lower Tavily's effective rate at volume, shown separately below).

At 10,000 queries a month, the same shapes cost $80–$20,000 (Tavily), $70 (Exa, shape A/B), and $50–$139.50 (Perplexity); at 1 million a month, $8,000–$2,000,000 (Tavily), $7,000 (Exa), and $5,000–$13,950 (Perplexity). The 62.5x spread inside Tavily's own research tier at 100,000 calls ($3,200 to $200,000) is larger than the difference between any two vendors on the simple-search shape.

Tavily's subscription tiers at volume

At exactly 100,000 basic searches a month, the Growth plan's $500 flat fee is 37.5% cheaper than the pay-as-you-go rate of $800, because the plan's per-credit rate falls to $0.005. Any volume below or above the plan's included credits changes this calculus: shortfall wastes the difference, and overage reverts to a higher rate on some plans.

Multiplier from retrieving more results or pages

Tavily's extract endpoint bills in blocks of 5 successful URL extractions. Formula: credits = pages ÷ 5 × 1 (basic) or 2 (advanced).

Retrieving 20 pages instead of 5 quadruples the extraction cost per query, a direct, linear multiplier rather than a discount for bulk retrieval within a single call. Exa's and Perplexity's per-request pricing does not itemize a separate per-result-count charge in the sources reviewed, so a request returning 5 versus 20 results costs the same on those two unless a separate content-extraction step is added.

Research-mode multiplier

Tavily's research endpoint spans 4 credits (mini model, simple query) to 250 credits (pro model, complex multi-step query), a 62.5x range driven by runtime query complexity that the caller does not directly control. This is the single largest cost-uncertainty factor in this comparison: a workload that occasionally triggers the top of that range can turn a forecasted $3,200-a-month bill into $200,000 without any change in query volume.

Keeping search separate from downstream generation

Tavily's and Exa's raw search and content-extraction prices above do not include a separate LLM call to turn results into a final answer; that cost belongs to whichever generation model the caller uses downstream, priced by the token. Perplexity's Sonar price already includes that synthesis, which is why its per-request fee is higher than a bare search call but may be cheaper than search-plus-separate-generation once the downstream model's token cost is added.

Sensitivity

1. Research query complexity. The gap between Tavily's 4-credit and 250-credit research calls is the dominant cost risk in this whole comparison.
2. Plan fit. Subscription tiers only pay off when usage lands close to the included credit allowance; under- or over-shooting it erodes the discount.
3. Extraction depth. Doubling pages retrieved per query roughly doubles or quadruples extraction cost, depending on basic versus advanced depth.
4. Synthesis versus raw results. Perplexity's all-in Sonar price should be compared against Tavily or Exa plus a real downstream generation cost, not against Tavily or Exa alone.

Budgeting traps

• Assuming research-mode cost is predictable. It is priced by runtime complexity, not by a fixed per-call rate.
• Comparing Sonar's per-request fee to a bare search price. Sonar includes generation; Tavily's and Exa's search endpoints do not.
• Ignoring extraction depth. Basic versus advanced extraction, and 5 versus 20 pages, both change cost linearly or worse.
• Sizing a subscription plan for average volume rather than peak. A plan that comfortably covers typical months can be blown out by a single high-research-complexity month.

What to ask before you buy

Ask each vendor for the actual credit or token consumption on a sample of your real queries, especially any research or deep-search calls, before committing to a subscription tier. Then price the downstream generation call separately when comparing a raw-search product against an all-in-one product like Sonar.

ARTICLE 9