> SpecStream is the live web data engine for teams and agents. This is the Markdown copy of https://specstream.ai/platforms. Every page, with what SpecStream does and costs: https://specstream.ai/llms.txt

# Search API for platforms: switch provider in one line | SpecStream

Compared on Nov 9, 2026.

# SpecStream for platforms

**Change your search provider in one line, at $5 per 1,000 searches.**

Your product gives its users web search and pays a provider for every call. SpecStream is the live web data engine, and its Search API is the most accurate in the world: 97.9 F-score on all 1,000 SimpleQA Verified questions, at $5 per 1,000 searches.

Moving costs one line. SpecStream answers the calls your code already sends to Google Custom Search, Tavily, Exa, Brave, Serper, Bing and SearXNG, in the same JSON. Prepay $1,000 or more and you get 20% more credit.

[Get an API key](https://specstream.ai/signup?src=alternatives-platforms&utm_source=alternatives&utm_campaign=platforms&returnTo=%2Fworkspace%2Fagents%3Ftab%3Dkeys)

New accounts get $10 of credit, no card. It covers 2,000 searches.

## SpecStream next to the search providers platforms use, from each company's own pages on Nov 9, 2026

|  | SpecStream | Exa | Tavily | Brave Search API | Parallel |
| --- | --- | --- | --- | --- | --- |
| Price per 1,000 searches | $5, with up to 10 results a search. $4.17 in effect with a prepayment of $1,000 or more, which adds 20% more credit | From $4 per 1,000 requests, and $7 per 1,000 for `auto` and `fast` search | $0.008 a credit on pay as you go | $5 per 1,000 requests on the Search plan | From $1 per 1,000 requests |
| SimpleQA Verified | 97.9 F-score on all 1,000 official questions, 978 correct | 91 on a sample of 100 questions, in Parallel's published run of September 9, 2026 | 92 to 94 on the same 100-question sample. Its own post reports 97.6% on SimpleQA | Publishes FreshQA on its page: 85.3% accuracy | 97 for Advanced and Basic and 94 for Fast, on a sample of 100 questions it ran on September 9, 2026 |
| WideSearch | 64.0 item F1 on all 200 official tasks | 53 to 55.9 on a sample of 100 tasks, in Parallel's published run | 47.9 to 55.9 on the same 100-task sample | Publishes an agent quality score of 14.89 from an AIMultiple test | 45.5 for Fast, 55.3 for Basic and 57.6 for Advanced, on a sample of 100 tasks |
| Moving from your provider | One changed address. SpecStream answers Google Custom Search, Tavily, Exa, Brave, Serper, Bing v7 and SearXNG calls on their own paths, in their own JSON | Its own request and reply shape | Its own request and reply shape | Its own request and reply shape | A migration guide that maps Exa, Tavily and Brave calls to the Parallel Search API |
| What is kept from a search | Nothing, on a key with zero retention: no request text and no answer after the reply, in the history or in a log line. You set it yourself on any key | Its privacy page says query data is used to improve its products, including by training and fine-tuning models | Its privacy page says it may use portions of your query data to improve its responses to future queries | Queries are never shared or trained on | Synchronous APIs retain no content once zero data retention is active |
| Who gets zero retention | Every account. One field on the key, `zero_retention`, at the same $5 per 1,000 | An Enterprise plan: its security page says to contact sales | Its privacy page says it keeps information for the period necessary to provide and improve the services | Enterprise customers | Not on by default. The Business plan includes it by signed agreement |

## Why platforms make SpecStream their search provider

### You move by changing one address

SpecStream speaks the API your product already calls. The path, the parameters and the JSON that comes back stay as they are, so the code that reads the reply does not change. This is a product on Tavily, before and after.

```
# Before
curl -X POST https://api.tavily.com/search \
  -H "Authorization: Bearer $TAVILY_API_KEY" -H "Content-Type: application/json" \
  -d '{"query": "Who founded Mistral AI and when?", "max_results": 5}'

# After
curl -X POST https://api.specstream.ai/tavily/search \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"query": "Who founded Mistral AI and when?", "max_results": 5}'
```

Seven routes are live today, each billed as one search on your key: Google Custom Search JSON API at `https://api.specstream.ai/customsearch/v1`, Tavily at `https://api.specstream.ai/tavily`, Exa at `https://api.specstream.ai/exa`, Brave at `https://api.specstream.ai/res/v1/web/search`, Serper at `https://api.specstream.ai/serper`, Bing v7 at `https://api.specstream.ai/v7.0/search` and SearXNG at `https://api.specstream.ai/searxng`.

Google ends its Custom Search JSON API on January 1, 2027. A product that runs on it [changes one line](https://specstream.ai/google-custom-search-api-alternative?utm_source=alternatives&utm_campaign=platforms) and pays the same $5 per 1,000, with no limit of 10,000 a day.

### Your users get the most accurate web search API in the world

On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score, with its method and each provider's published score beside it, is on the [benchmarks](https://specstream.ai/benchmarks?utm_source=alternatives&utm_campaign=platforms) page.

### Every answer is read from the live web

Each search is read from the web at the second of the call. There is no stored copy to go stale, so a page published an hour ago is in the answer your user gets.

### You decide what is kept, on each key

Create a key with `zero_retention` and a search made with it keeps no request text and no answer after the reply. It is not in the history, its answer is not reused for another call, and its log lines carry neither. The charge keeps the account, the key, the request ID, the price and the time.

```
curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"zero_retention": true}'
```

SpecStream SAS is a French company under EU law, and the service runs on IBM Cloud in Madrid, Spain.

### The paper is done the day you ask

Our [data processing agreement](https://specstream.ai/dpa?utm_source=alternatives&utm_campaign=platforms) is published. A platform that resells search gets a signed supplier data addendum the same day, and when you send your own paper with the same questions we sign it that day too. Write to hello@specstream.ai.

### $5 per 1,000 searches, and 20% more credit when you prepay

A search of up to 10 results costs $0.005, and each started block of 10 more results adds $0.006, up to 100 results a search. A call stopped by a `503` charges nothing, and an `Idempotency-Key` makes a retried call charge once.

Prepay $1,000 or more and you get 20% more credit: $1,000 buys $1,200 of use, which is $4.17 per 1,000 searches. Write to hello@specstream.ai and the payment link comes back the same day. Credit you buy never expires.

### Search, page reading and finished rows from one supplier

The same key reads up to 20 pages or documents a call with the [Extract API](https://specstream.ai/extract-api?utm_source=alternatives&utm_campaign=platforms) at $0.80 per 1,000 pages, and a page with no text is not charged. [Deep Search](https://specstream.ai/deep-search?utm_source=alternatives&utm_campaign=platforms) returns every match for a request in plain words as structured rows, at 5¢, 15¢ or $1 a row, with a source URL on every value. Your product gets search, pages and data from one contract and one bill.

## Who moves to SpecStream

### A product that bills or meters web search

You sell searches to your users and name your provider in your docs. At $5 per 1,000, or $4.17 with a prepayment, your margin on each search is known before you ship, and your users get the top score on the official sets.

### A model host with no search tool

OpenAI and Anthropic charge $10 per 1,000 searches for their own search tools. Give every model you host the same web search at $5 per 1,000, through the API or the MCP server at `https://api.specstream.ai/mcp`.

### A private or European assistant

Your promise to your users is that their questions stay theirs. A zero retention key keeps no request text and no answer, the supplier is a French company, and the servers are in Madrid.

### Any product on Google's Custom Search JSON API

Google ends the API on January 1, 2027. SpecStream answers the same call at `https://api.specstream.ai/customsearch/v1` in Google's JSON, at the same $5 per 1,000 and with no daily limit.

## How SpecStream scores

The same numbers as the benchmarks page, from the official tasks and scorers.

ResultsWideSearchCostSimpleQA Verified

|  | SpecStream | ![](https://specstream.ai/images/benchmarks/parallel.svg)Parallel | ![](https://specstream.ai/images/benchmarks/exa.svg)Exa | ![](https://specstream.ai/images/benchmarks/tavily.svg)Tavily | ![](https://specstream.ai/images/benchmarks/perplexity.svg)Perplexity |
| --- | --- | --- | --- | --- | --- |
| **Web research**WideSearch, GPT-5.6 Luna | 60.9 | 45.5 | 53.0 | 47.9 | 47.0 |
| **Factual accuracy**SimpleQA Verified, GPT-5.6 Luna | 96 | 94 | 91 | 94 | 94 |
| **Factual accuracy**SimpleQA Verified, GPT-5.6 Sol | 99 | 97 | 91 | 92 | 95 |

## How to start

1\. Create a key under Agents & API in the workspace. A new account gets $10 of credit, no card, which covers 2,000 searches. Turn on `zero_retention` on the key when your users' searches must leave no trace.

2\. Point your client at the route for the API it speaks today, and send the SpecStream key where the old key went. Run your own questions and compare the answers.

3\. Write to hello@specstream.ai for the supplier data addendum and the prepayment link. Both come back the same day.

[Get an API key](https://specstream.ai/signup?src=alternatives-platforms&utm_source=alternatives&utm_campaign=platforms&returnTo=%2Fworkspace%2Fagents%3Ftab%3Dkeys)

## Frequently asked questions

Something we didn't answer? [Talk to a human](https://specstream.ai/talk-to-a-human.md)

How much does the SpecStream Search API cost for a platform?

$5 per 1,000 searches of up to 10 results, which is $0.005 a call. A prepayment of $1,000 or more adds 20% more credit, so 1,000 searches cost $4.17 in effect. A call stopped by a `503` is not charged, and credit you buy never expires.

Can I switch from Tavily, Exa, Brave, Serper, Bing or Google Custom Search without changing my code?

Yes. You change one address and the key. SpecStream answers each of those calls on its own path under `https://api.specstream.ai`, in the same JSON shape, so the code that reads the reply stays as it is. Each call is billed as one search.

What does SpecStream keep from my users' searches?

Nothing, on a key with zero retention. A search made with such a key keeps no request text and no answer after the reply, in the history or in a log line. The charge keeps the account, the key, the request ID, the price and the time.

How accurate is the SpecStream Search API?

It scores 97.9 F-score on all 1,000 SimpleQA Verified questions, with 978 correct, and 64.0 item F1 on all 200 WideSearch tasks. Both are the full official sets, and the method is on the [benchmarks](https://specstream.ai/benchmarks?utm_source=alternatives&utm_campaign=platforms) page.

Do you sign a DPA or a supplier data addendum?

Yes, the same day. The [data processing agreement](https://specstream.ai/dpa?utm_source=alternatives&utm_campaign=platforms) is published, and a platform that resells search gets a signed supplier data addendum the day it asks at hello@specstream.ai.

Where does SpecStream run, and who is the company?

SpecStream SAS is a French company in Narbonne, under EU law. The service runs on IBM Cloud in Madrid, Spain.

What else can my product buy on the same key?

Page reading and finished data. The Extract API reads up to 20 pages or documents a call at $0.80 per 1,000 pages, and Deep Search returns every match for a request as structured rows at 5¢, 15¢ or $1 a row.

### Sources

SpecStream wrote this page. Each provider's figure comes from that company's own page, or from the page named in the cell.

1.  [Exa pricing](https://exa.ai/pricing)exa.ai/pricing, read Nov 9, 2026
2.  [Exa security](https://exa.ai/docs/reference/security)exa.ai/docs/reference/security, read Nov 9, 2026
3.  [Exa privacy policy](https://exa.ai/privacy)exa.ai/privacy, read Nov 9, 2026
4.  [Tavily pricing](https://www.tavily.com/pricing)tavily.com/pricing, read Nov 9, 2026
5.  [Tavily privacy policy](https://www.tavily.com/privacy)tavily.com/privacy, read Nov 9, 2026
6.  [Tavily, SealQA and SimpleQA post](https://www.tavily.com/blog/tavily-ranks-1-on-sealqa-and-simpleqa)tavily.com/blog/tavily-ranks-1-on-sealqa-and-simpleqa, read Nov 9, 2026
7.  [Brave Search API](https://brave.com/search/api/)brave.com/search/api/, read Nov 9, 2026
8.  [Parallel pricing](https://parallel.ai/pricing)parallel.ai/pricing, read Nov 9, 2026
9.  [Parallel benchmarks](https://parallel.ai/benchmarks)parallel.ai/benchmarks, read Nov 9, 2026
10.  [Parallel, migrate to Parallel](https://docs.parallel.ai/search/migrate-to-parallel)docs.parallel.ai/search/migrate-to-parallel, read Nov 9, 2026
11.  [OpenAI API pricing](https://developers.openai.com/api/docs/pricing)developers.openai.com/api/docs/pricing, read Nov 9, 2026
12.  [Anthropic pricing](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool)platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool, read Nov 9, 2026
13.  [Google Custom Search JSON API overview](https://developers.google.com/custom-search/v1/overview)developers.google.com/custom-search/v1/overview, read Nov 9, 2026
14.  [SpecStream benchmarks](https://specstream.ai/benchmarks)specstream.ai/benchmarks, read Nov 9, 2026
15.  [SpecStream for agents](https://specstream.ai/agents.md)specstream.ai/agents.md, read Nov 9, 2026
16.  [SpecStream data processing agreement](https://specstream.ai/dpa)specstream.ai/dpa, read Nov 9, 2026