Web Search Tool Cost Calculator
Grounded search costs $10-35 per 1k queries — price your agent's search habit accurately.
content tokens at GPT-5.4 input rate; retrieved pages typically inject 3k–10k tokens per search.
per month on OpenAI web search — $152.20 in search fees + $228.30 in content-token costs at 15,220 searches/month.
- Search fee share
- 40%
- Token fee share
- 60%
- Billable searches
- 15,220
- Cost per search (all-in)
- $0.03
search fees ■ vs content-token cost split · rates verified 2026-06-11
How it works
Web search is the most commonly enabled agent tool and the most commonly unbudgeted one. This calculator splits the real cost into its two parts — the per-query search fee and the content-token cost of stuffing retrieved pages into your model's context — because the second part is usually the bigger number and almost nobody prices it.
The fee schedules: OpenAI bills $10 per 1,000 web-search tool calls, with retrieved content billed as ordinary input tokens at the calling model's rate (this tool uses GPT-5.4's $2.50/M as the representative rate). Anthropic also bills $10 per 1,000 searches plus content tokens at the model rate (Sonnet 4.5's $3/M here). Gemini's Grounding with Google Search includes 5,000 free grounded prompts per month shared across Gemini 3 models, then charges $14 per 1,000 queries — the calculator deducts the free tier before computing fees.
The split bar is the point of this tool. At a typical 6,000 retrieved tokens per search, the token cost is already larger than the search fee on every provider, and it scales with how verbose the retrieved pages are, not with how many searches you run. An agent that searches once and then carries 8,000 tokens of page content through ten more turns pays input-token rent on that content every single turn — a cost this calculator approximates only for the initial injection, so treat the total as a floor.
One modeling note documented in the source: OpenAI's non-reasoning search-preview variant prices at $25 per 1,000 calls with content tokens free, which can win for content-heavy searches on expensive models. The standard tool is what agentic stacks actually use, so that is what the calculator prices.
All rates were verified on 2026-06-11 against the OpenAI, Anthropic and Google pricing pages, and the verification date is embedded as a documented constant in the tool source. Monthly volume uses a 30.44-day month. For measured search-tool spend from your real agent traffic — including the repeated context rent this estimate cannot see — FORG breaks it out per session automatically. The share link preserves your exact inputs for budget reviews.
Frequently asked questions
How much does the web search tool cost per query?
OpenAI charges $10 per 1,000 web-search tool calls and Anthropic charges $10 per 1,000 searches — a flat penny per query on both. Gemini's Grounding with Google Search gives 5,000 free grounded prompts per month shared across Gemini 3 models, then bills $14 per 1,000 queries. On top of all three, the retrieved page content is billed as input tokens at the calling model's normal rate.
Why do content tokens matter more than the search fee?
A single search typically injects 3,000-10,000 tokens of retrieved page content into the model's context, billed at the model's input rate. At 6,000 tokens on a $2.50/M model that is $0.015 per search — already 1.5× the $0.01 search fee. Agents that search in loops pay the token tax on every subsequent turn too, because the content stays in context.
What about the OpenAI search-preview pricing at $25/1k?
OpenAI's non-reasoning search-preview variant charges $25 per 1,000 calls but includes the retrieved content tokens for free, which can win for very content-heavy searches on expensive models. This calculator models the standard $10/1k tool where content bills at model rates, since that is the common path for agentic workloads; the constant in the source documents both.
How does Gemini's free grounding tier work?
Google includes 5,000 grounded prompts per month at no charge, shared across the Gemini 3 model family, before the $14/1k rate applies. Gemini 2.5 Pro has a different scheme — 1,500 free requests per day, then $35/1k. The calculator deducts the monthly free tier from your volume before computing fees, which makes Gemini effectively free for many low-volume agents.
FORG tracks this automatically across every agent session — live cost attribution, budgets, and alerts.
Start tracking with FORG