How much does Claude's web search tool cost per 1,000 searches?

Anthropic charges $10 per 1,000 web searches for the Claude web search tool, billed on top of normal token costs. The retrieved page content is also added to your prompt and billed as input tokens at the model's input rate — at Claude Sonnet's $3/1M that is often the larger half of the bill once you fetch several thousand tokens per search.

Does OpenAI charge for web search separately from tokens?

Yes. The OpenAI Responses API web search tool has a per-call fee that varies by model and search-context size (low, medium or high), and the content the tool retrieves is billed separately as input tokens. So a single answered question can incur a tool fee plus several thousand input tokens of fetched text. This calculator models both.

How much does Gemini grounding with Google Search cost?

Gemini grounding with Google Search gives a free daily quota of grounded prompts (1,500/day on the paid tier at last check), then charges roughly $35 per 1,000 grounded prompts above it, plus token costs. The calculator subtracts the free quota (1,500 × 30 days = 45,000/month) before applying the fee, so it never over-bills the free tier.

Why is my LLM bill higher when web search is enabled?

Two reasons stack: the per-search tool fee, and the page content each search injects into your prompt as input tokens. At 2 searches per request and 8,000 retrieved tokens per search, every answer adds 16,000 input tokens before your own prompt — usually a bigger line item than the search fee itself. The breakdown bar shows the split.

Is hosted web search cheaper than building my own RAG pipeline?

It depends on volume and freshness needs. Hosted web search has near-zero setup and always-fresh results, but the per-search fee plus retrieved tokens scale linearly with traffic. A self-hosted retrieval/embeddings pipeline has upfront cost but cheaper marginal queries. For a self-hosted comparison, use the RAG cost calculator; this tool prices only the hosted web-search add-on.

Are retrieved search results billed as input tokens?

Yes, on every provider modelled here. The text a search fetches is appended to your context window and charged at the input-token rate, in addition to the per-call search fee. That double charge is the whole reason this calculator exists — pricing pages quote the two costs in different places, so the real total is easy to under-estimate.

Can I change the rates if my pricing is different?

Yes. Selecting a model auto-fills the search fee, input price, output price and free quota, but every field is editable. Override them to model an enterprise discount, a different OpenAI search-context tier, or a future price change. The comparison table always uses each provider's published rates so it stays a fair like-for-like.

Are the prices in this calculator current?

Anthropic's $10/1,000-searches web-search fee and its token rates (Sonnet 4.6 $3/$15, Opus 4.8 $5/$25, Haiku 4.5 $1/$5 per 1M) are cross-checked against the Anthropic docs. OpenAI, Google and Perplexity rows are published list prices that change without notice and are flagged in the tool. All were last verified 2026-06-21.

AI · Developer cost

AI Web Search / Grounding Cost Calculator

Giving an LLM live web access costs more than the per-search fee everyone quotes — the page content it fetches is billed as input tokens too. Estimate the real monthly bill, split into search fees vs retrieved tokens, across Claude, GPT, Gemini and Sonar. No signup, no API key.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jun 21, 2026

Web search / grounding cost

Rate verified

Provider & model

Currency

Requests / month

API requests you make in a month.

Searchs / request

Average per request. Decimals allowed (e.g. 2.5).

Retrieved tokens / search

Fetched page content injected per search.

Prompt tokens / request

Your own input tokens, excluding retrieved content.

Output tokens / request

Average generated tokens per request.

Rates — auto-filled from Claude Sonnet 4.6, editable

$ / 1,000 searchs

Per-call tool fee.

$ / 1M input tokens

Prices retrieved + prompt tokens.

$ / 1M output tokens

Prices generated tokens.

Free searchs / day

Grounded free tier, if any.

Total / month

$2,508.00

Claude Sonnet 4.6

Search / grounding fee

$600.00

60,000 searchs, no free tier

Retrieved-token cost

$1,440.00

57.42% of bill

Per request

$0.0836

60,000 searchs total

Cost breakdown

Search / grounding fee

$600.00 (23.92%)

Retrieved-content tokens

$1,440.00 (57.42%)

Prompt + output tokens

$468.00 (18.66%)

Same workload, every provider

Model	Search fee	Retrieved	Prompt+out	Total / mo
Gemini 3 Flash (grounding)cheapestlist price	$525.00	$144.00	$70.80	$739.80
Perplexity Sonarlist price	$300.00	$480.00	$60.00	$840.00
Claude Haiku 4.5	$600.00	$480.00	$156.00	$1,236.00
Gemini 3 Pro (grounding)list price	$525.00	$600.00	$285.00	$1,410.00
GPT-4o mini (web search)list price	$1,500.00	$72.00	$19.80	$1,591.80
Claude Sonnet 4.6selected	$600.00	$1,440.00	$468.00	$2,508.00
GPT-4o (web search)list price	$1,500.00	$1,200.00	$330.00	$3,030.00
Claude Opus 4.8	$600.00	$2,400.00	$780.00	$3,780.00

Each provider priced on its own published rates and free quota — your edited rates apply only to the selected model above. Sorted cheapest-first; Gemini's daily grounded free tier (1,500/day × 30 days) is applied before its fee.

A hosted web-search / grounding tool bills a per-call fee AND the input tokensof the page content it fetches into your prompt. This prices both. Anthropic's $10/1,000-searches fee and its token rates are cross-checked; OpenAI, Google and Perplexity rows are published list prices that change without notice. Rates last verified 2026-06-21. Full sources are listed below the calculator.

How it works

A hosted “web search” or “grounding” tool has two costs that provider pricing pages list in different places. The first is the per-call fee — Anthropic bills $10 per 1,000 searches, Google bills per 1,000 grounded prompts after a free daily quota, and OpenAI bills per call by search-context size. The second, and usually larger, is the retrieved content: every search fetches page text that is appended to your prompt and billed as input tokens at the model's input rate. The calculator prices both and shows the split.

The math, per provider, is:

Total searches = requests × searches_per_request. Searches are an average, so this may be fractional.
Chargeable searches = max(0, total_searches − free_per_day × 30). Providers with a daily free grounded quota (Gemini) have it subtracted first; the chargeable count is floored at zero so the free tier never over-bills.
Search fee = chargeable_searches ÷ 1,000 × fee_per_1k.
Retrieved-content cost = total_searches × retrieved_tokens ÷ 1,000,000 × input_$/1M — the page text billed as input tokens.
Base token cost = your own prompt input plus generated output: requests × (prompt_tok × input_$/1M + output_tok × output_$/1M) ÷ 1,000,000.
Total monthly cost = search fee + retrieved-content cost + base token cost.

The comparison table re-runs this identical formula for every seeded provider using that provider's own fee, token prices and free quota, then sorts cheapest-first. Your edited rates apply only to the selected model so the comparison stays a fair like-for-like. Every headline total is cross-checked by an independent segment-sum (search + retrieved + prompt + output), which agrees to the cent with the grouped total. All prices are static, human-verified constants dated by LAST_VERIFIED — no live pricing calls.

Worked examples

Claude Sonnet 4.6 — research assistant

Inputs: 30,000 requests/mo · 2 searches/request · 8,000 retrieved tok/search · 1,200 prompt tok · 800 output tok.
Rates: $10/1k searches · $3/1M in · $15/1M out · no free quota.
Total searches: 30,000 × 2 = 60,000 → search fee = 60 × $10 = $600.00
Retrieved tokens: 60,000 × 8,000 = 480,000,000 → 480 × $3 = $1,440.00
Base input: 30,000 × 1,200 = 36,000,000 → 36 × $3 = $108.00
Output: 30,000 × 800 = 24,000,000 → 24 × $15 = $360.00
Total = 600 + 1,440 + 108 + 360 = $2,508.00/month
Search fee is only ~24%; retrieved tokens are ~57% — the hidden half.

Gemini 3 Flash grounding — free quota applied

Same workload, but $35/1k grounded prompts, free 1,500/day, $0.30/1M in, $2.50/1M out.
Total grounded prompts: 60,000 · free = 1,500 × 30 = 45,000
Chargeable = max(0, 60,000 − 45,000) = 15,000 → fee = 15 × $35 = $525.00
Retrieved: 480 × $0.30 = $144.00 · base input: 36 × $0.30 = $10.80
Output: 24 × $2.50 = $60.00
Total = 525 + 144 + 10.80 + 60 = $739.80/month
Skip the quota and you'd over-bill by 45 × $35 = $1,575.

At the free-quota boundary (edge case)

Gemini grounding, 22,500 requests/mo × 2 = 45,000 grounded prompts.
Free monthly quota = 1,500 × 30 = 45,000.
Chargeable = max(0, 45,000 − 45,000) = 0 → search fee = $0.00
One more request: 22,501 × 2 = 45,002 → chargeable 2 → fee = 0.002 × $35 = $0.07
Token costs still apply; only the search fee is absorbed by the free tier.

Frequently asked questions

Sources & references

Prices were last cross-checked against the cited sources on 2026-06-21. Anthropic rates are verified against the Anthropic web-search and pricing docs; OpenAI, Google and Perplexity rows are published list prices that change without notice — re-check the provider page before relying on a figure. Rates are USD; the rupee view uses an indicative, editable USD→LKR rate.

Related tools

LiveAI

AI Chatbot Cost Calculator

Estimate the monthly API cost of a multi-turn AI chatbot across Claude, GPT, and Gemini. Models the quadratic context re-sending that single-call calculators miss, with and without prompt caching, in USD and LKR.

Open tool

LiveAI

Reasoning Token Cost Calc

Estimate the true cost of reasoning-model API calls by accounting for the hidden reasoning/thinking tokens that o-series, Claude, and Gemini bill at output rates. See per-call and monthly cost in USD and LKR, plus how much more it is than a naive estimate.

Open tool

LiveAI

AI Agent Cost Calculator

Estimate the real per-run, daily, and monthly cost of a multi-step LLM agent across Claude, GPT, and Gemini. Models the context accumulation single-call calculators miss — each tool result is re-sent every step — with caching, in USD and LKR.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or want to suggest an improvement?

Email me at [email protected] — most fixes ship within 24 hours.