String won the measured fetch comparison. Nimble has the most important adapter caveat in the field. On the open Web Data Frontier Benchmark, String returned verified content on 97.0% of 495 requests and Nimble returned 66.3%. String's failure-aware latency score was 9.98 seconds against Nimble's 18.41. But the August run called Nimble's legacy v1 Extract endpoint with render: false; Nimble's current v2 API offers browser rendering and automatic driver selection. Treat 66.3% as the result of that exact non-rendered path, not as Nimble's current ceiling.
TL;DR
render: false, while new integrations should use v2| String | Nimble | |
|---|---|---|
| Product type | Fetch, search, sitemap, and remote browser | Extract, search, map, crawl, agents, templates, media, and proxies |
| Benchmark success rate | 97.0% (480/495) | 66.3% (328/495), legacy v1 non-rendered path |
| Latency score | 9.98s, second of 15 | 18.41s, 13th of 15 |
| Per-target outcome | 92 perfect, 5 partial, 2 zero | 58 perfect, 12 partial, 29 zero |
| Anti-bot highlights | 100% Cloudflare, Kasada, and Fastly | 82.2% PerimeterX, 83.3% AWS WAF, 100% Fastly |
| Pricing basis | Plan plus successful-request usage; credits roll over | $1 per 1,000 successful Extract URLs, no commitment |
| Free evaluation | First 5,000 requests | 5,000 free pages, no card required |
| Agent tooling | Hosted MCP and HTTP API | Hosted MCP, Web Search Agents, CLI, Python, Node, and Go SDKs |
| Where it wins | Measured protected-page coverage and latency | Two targets, flat extraction price, product breadth, enterprise controls |
We built the Web Data Frontier Benchmark and sell the product that ranks first. The defense against that conflict is reproducibility. The 99 targets and pass criteria, provider adapters, and August results JSON are public. A request passes only when its body contains an expected marker from the real page. A captcha or empty shell with HTTP 200 fails.
The Nimble adapter called POST https://sdk.nimbleway.com/v1/extract with the URL and render: false. Its own source comment describes that as raw HTML over a residential Nimble IP, with no headless browser. It did not use JavaScript rendering, browser actions, or Nimble's current automatic driver flow.
That qualification is not small. Nimble's current v2 introduction says new integrations should target v2, and the current Extract API supports auto, VX6, VX8, VX10, VX12, Pro variants, JavaScript rendering, and browser actions. An auto-driver v2 rerun could score higher. Until that run exists, this page reports the measured result and refuses to present it as Nimble's product ceiling.
The harness is open to a corrected adapter. A Firecrawl engineer used that route in benchmark pull request 6; the next run improved Firecrawl's result by 7.7 points. Nimble can do the same.
Pricing and product claims were checked against live first-party pages on August 31, 2026. Performance figures are from the August 11 run.
Nimble's current platform is much broader than the endpoint tested in August. The v2 API includes URL extraction, web search, Web Search Agents, extraction templates, mapping, crawling, SERP retrieval, media downloads, scheduled jobs, and async task and batch APIs. Its Web API combines Nimble's browser drivers with its residential proxy network and can deliver raw or structured data.
The Extract endpoint can return HTML and Markdown, parse structured output, choose a driver, render JavaScript, run browser actions, set location and device parameters, capture network traffic, and manage sessions. Current docs list a default 83 queries per second for the VX driver family.
Nimble also ships a hosted MCP server with tools for search, extract, map, crawl, extraction templates, and research agents. String is narrower: one fetch endpoint with automatic escalation, plus search, sitemap crawling, a remote browser over CDP, and MCP.
The benchmark sent 15 providers the same 99 URLs five times each. Most targets are protected retail, travel, news, real-estate, and social pages.
| Provider | Successful requests | Success rate | Latency score |
|---|---|---|---|
| String | 480/495 | 97.0% | 9.98s |
| Scrapfly | 406/495 | 82.0% | 18.35s |
| Context.dev | 392/495 | 79.2% | 12.68s |
| Firecrawl | 389/495 | 78.6% | 9.21s |
| Bright Data | 386/495 | 78.0% | 26.14s |
| Nimble | 328/495 | 66.3% | 18.41s |
Nimble ranked ninth for success and 13th on the latency score. The score uses each target's p75 successful latency. A complete miss inherits that target's p75 from providers that did return content, or the 90-second timeout when nobody did. Fast failures do not improve it.
Again, those ranks belong to Nimble's v1 render: false adapter. They are useful for teams considering that exact cheap, non-rendered path. They do not answer what v2 driver: auto or a rendered VX driver would have done.
Nimble passed Neiman Marcus and Booking.com on all five attempts. String passed each four times. They were the only two targets where Nimble beat String outright.
The industry view has other strengths. Nimble passed every request in the news and finance category and every developer and research request. It reached 95% across social platforms, 85.7% in real estate, and 85% in tickets and events. Against named anti-bot systems, it posted 82.2% on PerimeterX, 83.3% on AWS WAF, and 100% on the single Fastly target.
Those results show why the adapter caveat cuts both ways. A non-rendered path already handled all news, finance, developer, and research requests and found two pages String missed once. Teams targeting those surfaces should test Nimble directly.
| Outcome across 99 sites | String | Nimble |
|---|---|---|
| Passed all 5 attempts | 92 | 58 |
| Passed 1 to 4 attempts | 5 | 12 |
| Passed none | 2 | 29 |
Nimble's non-rendered path returned nothing on 29 sites. The list includes Safeway, Trustpilot, Best Buy, Crunchbase, Apartments.com, Lululemon, Home Depot, Expedia, AutoZone, Yelp, Hyatt, Saks Fifth Avenue, G2, AutoTrader, StockX, ZipRecruiter, Zara, American Airlines, Cars.com, Capterra, ASDA, and GOAT. String passed 25 of the 29 on all five attempts, Mouser and Coupang four times, and also failed Temu and Skyscanner.
The anti-bot columns show the same shape. Nimble passed 34.7% of Cloudflare requests against String's 100%, 44.0% of Akamai against 98.0%, and 71.6% of DataDome against 97.9%. Its strongest multi-site named wall was AWS WAF at 83.3%, against String's 96.7%.
Nimble's current API pricing charges $1 per 1,000 successful Extraction Tool URLs, with no commitment. Only successful requests bill. The free trial includes 5,000 pages with no card. Search costs $1.10 per 1,000 lite inputs or $5 per 1,000 fast inputs. Web Search Agents range from $0.025 to $2 per task by effort. Residential proxy traffic is $5.30 per GB.
String charges a monthly plan plus successful-request usage. Starter is $20 per month, then $0.40 per 1,000 standard fetches or $4 per 1,000 premium fetches. Growth is $100, then $0.20 or $2. Blocks are free and credits roll over.
At 100,000 calls against the benchmark target mix, the cheapest published String plan for each path gives this model:
| Assumption | Monthly cost | Cost per 1,000 usable pages | |
|---|---|---|---|
| Nimble Extract | 66,300 successful URLs at $1/1k; failures free | $66.30 | $1.00 |
| String Starter, standard fetch | 97,000 successes at $0.40/1k, plus $20 plan | $58.80 | $0.61 |
| String Growth, premium fetch | 97,000 successes at $2/1k, plus $100 plan | $294.00 | $3.03 |
This is a model, not a bill. String chooses a path based on what a page needs, so production traffic mixes rates. Nimble's current price page lists one $1 extraction rate rather than a separate rendered surcharge. Nimble is cheaper when most String requests would use the premium path. String is cheaper when most use its standard path. Both vendors exclude failed requests from billing, so retries do not create the same cost penalty as call-based competitors.
Nimble's advantage grows outside one-page fetches. Its Search API can retrieve and summarize results with domain, date, locale, and focus controls. Map discovers a site's URLs. Crawl retrieves multiple pages asynchronously. Extract Templates add reusable structured parsers, and Web Search Agents run longer cited research tasks. The platform offers batch and async extraction plus its own residential proxy API.
For agent clients, Nimble's hosted MCP server exposes that surface through Streamable HTTP. It also offers a CLI and Python, Node, and Go SDKs. String's MCP is centered on fetch, search, and sitemap tools.
Nimble also publishes a stronger enterprise-control page. Its compliance documentation states SOC 2 Type II, zero data retention, GDPR and CCPA coverage, standard contractual clauses, encryption, RBAC, and audit trails. Choose Nimble when those controls and the surrounding data platform outweigh the measured fetch gap.
String was better on the measured fetch task: 97.0% success against 66.3%, with a faster failure-aware latency score. Nimble is broader and cheaper than String's premium fetch path. Its benchmark result used legacy v1 with rendering disabled, so a current v2 auto-driver evaluation is still needed.
Nimble Extract is $1 per 1,000 successful URLs with no commitment. String Starter standard fetch is $0.40 per 1,000 plus $20 per month, while Growth premium fetch is $2 per 1,000 plus $100. String is cheaper for mostly standard traffic; Nimble is cheaper for mostly premium String traffic.
No. Nimble's pricing page says only successful API requests are charged. String also bills only requests that return content.
The public adapter called legacy POST /v1/extract with render: false. It returned raw HTML over a residential Nimble IP and did not use a headless browser, JavaScript rendering, browser actions, or current v2 automatic driver selection.
Sometimes on the tested non-rendered path. It passed 34.7% of requests across 15 Cloudflare targets, while String passed 100%. Nimble's current rendered and automatic drivers were not tested in the August run.
Yes. Its hosted MCP server exposes search, extract, map, crawl, templates, and Web Search Agent operations to compatible clients through Streamable HTTP.
Use Nimble when you need one web-data platform for extraction, search, research agents, maps, crawls, templates, proxies, and enterprise controls, or when its flat $1 successful-extract rate beats your String path mix. Use String when verified page retrieval is the main job.
Run the open-source benchmark with your own keys. The results JSON, target markers, and provider adapters are public. For Nimble, update the adapter to v2 and choose the driver configuration you actually plan to buy.
