// open source reference
Web search pricing in the APIs answer-engine-benchmark calls
answer-engine-benchmark asks OpenAI, Gemini, Claude and Perplexity questions with web search on. Each API charges for that search per call, on top of tokens. This page lists the current price and every change we could see on each vendor's own pricing page.
Price on 8 Nov 2026
| Vendor | Tool | Price |
|---|---|---|
| OpenAI | Web search (all models) | $10 per 1,000 calls, plus search content tokens at model rates |
| Grounding with Google Search, Gemini 3 and newer | 5,000 free requests a month, then $14 per 1,000 requests | |
| Anthropic | Web search tool | $10 per 1,000 searches, plus token costs |
| Perplexity | Agent API web_search (standard) | $0.0025 per invocation ($2.50 per 1,000) |
Change log
A change appeared between the last archived copy with the old text and the first with the new one. Both copies are linked. The Wayback Machine does not capture every day, so the window is as exact as the archive allows.
| Appeared | Vendor and tool | Old | New | Sources |
|---|---|---|---|---|
| Between and | Perplexity, Agent API web_search, Fast Search addedAll Agent API models | Not listed | $0.001 per invocation for search_type "fast"Standard web_search stayed at $0.0025. The benchmark sends no search_type. | Pricing pageArchived beforeArchived after |
| Between and | Perplexity, Agent API web_searchAll Agent API models, sonar included | $0.005 per invocation | $0.0025 per invocationThe price halved. | Pricing pageArchived beforeArchived after |
| Between and | Google, Grounding with Google SearchGemini 3.x, gemini-3.5-flash included | $14 per 1,000 search queries | $14 per 1,000 requestsThe unit was renamed from search queries to requests. The price did not change. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search, openai.com pricing cardNot stated on the card | $25 per 1,000 calls | $10 per 1,000 callsThe card says search content tokens are free. The developer page says they are billed at model rates. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search, openai.com pricing cardNot stated on the card | $10 per 1,000 calls | $25 per 1,000 callsA redesigned page with one card. OpenAI's developer pricing page showed $10 for web search on all models the whole time. | Pricing pageArchived beforeArchived after |
| Between and | Google, Grounding with Google SearchGemini 3 | $14 per 1,000 search queries, "Coming soon" | $14 per 1,000 search queries"Coming soon" was removed. The price did not change. | Pricing pageArchived beforeArchived after |
| Between and | Perplexity, Agent API web_searchAll Agent API models, sonar included | Not listed | $0.005 per invocationThe pricing page moved in this window. The older address showed only the Sonar API's request fees. | Pricing pageArchived beforeArchived after |
| Between and | Google, Grounding with Google SearchGemini 3 | Not listed | $14 per 1,000 search queries, marked "Coming soon"The free allowance later became 5,000 a month. Google's page said each search query a request triggers is charged. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search (the non-preview tool)All models | Not listed | $10 per 1,000 callsThis is the web_search tool the benchmark sends. Preview tools stayed at $25 (non-reasoning) and $10 (reasoning). | Pricing pageArchived beforeArchived after |
| On | OpenAI, Web search toolGPT-5, GPT-5-mini, GPT-5-nano added to the reasoning row | Not listed | $10 per 1,000 callsgpt-5-mini is the benchmark's default OpenAI model. | Pricing pageArchived beforeArchived after |
| Between and | Anthropic, Web search toolAll Claude models | Not listed | $10 per 1,000 searchesThe earlier page said only that web search charges per search performed. The price has not changed since. | Pricing pageArchived beforeArchived after |
| Appeared | Vendor and tool | Old | New | Sources |
|---|---|---|---|---|
| On | Google, Grounding with Google SearchGemini 2.x | $35 per 1,000 requests | $35 per 1,000 grounded promptsThe unit was renamed. The price did not change. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search toolReasoning models: o3, o4-mini, o3-pro, deep research | Not listed | $10 per 1,000 callsSearch content tokens billed at the model's rate. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search toolgpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini | $25 to $50 per 1,000 calls, by search context size | $25 per 1,000 callsSearch content tokens free. | Pricing pageArchived beforeArchived after |
| Between and | OpenAI, Web search toolgpt-4o and gpt-4o-mini | Not listed | $25 to $50 per 1,000 calls, by search context sizegpt-4o: $30, $35 or $50. gpt-4o-mini: $25, $27.50 or $30. Low, medium or high context. | Pricing pageArchived beforeArchived after |
| Between and | Google, Grounding with Google SearchGemini 1.5 | Not listed | $35 per 1,000 grounding requestsPaid tier only. The free tier read "Not available". | Pricing pageArchived beforeArchived after |
How this is checked
We read each vendor's current pricing page, then its snapshots in the Wayback Machine, and narrowed every difference in the web search line down to two neighbouring snapshots. Only text on the vendor's own pricing page counts. Announcements, changelogs elsewhere and third-party trackers do not. A price that changed and changed back between two snapshots would not show here.
The benchmark's defaults are gpt-5-mini, gemini-3.5-flash, claude-sonnet-5 and perplexity/sonar on Perplexity's Agent API. Rows for other models on the same pages are in the second table. Perplexity's older Sonar API request fees are left out, because the benchmark does not call that API.
Synapse rechecks this page every month. If you see a price here that no longer matches the vendor's page, tell us and we will correct it and date the correction.