Firecrawl Review (September 2026): Developer Index, ChatGPT Plugin, and the 4.5 Revisited

Four things shipped in August that the July review never saw, and one of them changes what a coding agent buys Firecrawl for. The Developer Index (Aug 20) gives an agent a semantic index over 70M+ real GitHub artifacts. The Research Index went fully free and added 41M+ life-sciences papers (Aug 13). The parsing stack behind /parse got open-sourced (Aug 6). And Firecrawl became an official ChatGPT plugin (Aug 5). The July rating was 4.5. It moves to 4.6, and the reason is narrow: the same-priced product now does a job for coding agents that no raw scraper does.

Here is the decision up front. The tier you buy still turns on credit burn, not features. Start keyless or on Hobby, jump to Standard once your steady-state burn crosses roughly 6,000 credits a month, and treat the Developer Index as the feature that justifies picking Firecrawl over a plain HTML scraper if you are building a coding agent. What August did not do is change the pricing, so anyone who priced this out in July can keep their numbers, with one caveat on the dollar amounts that I will get to.

We pulled the current pricing page, read all four August launch posts against the docs, and checked where the new endpoints leave holes in the credit math. The short version: broader product, same subscription, one new budgeting unknown.

What shipped in August 2026

Five announcements landed between early August and the connector that preceded them in late July. Each one moves who Firecrawl is for, so here is what each does and where the source is.

Developer Index (Aug 20). This is the one that matters for the target reader. A coding agent spends a large share of its tool calls hunting for the repo that implements an idea, the docs page that answers a question, or the PR where a bug got fixed. The Developer Index indexes 70M+ artifacts (READMEs, issues, pull requests, OpenAPI specs, skills, and external documentation), refreshed continuously with most sources updated daily, and returns ranked developer results with the matching passages so the agent can act without a second scrape (firecrawl.dev/blog/developer-index-launch, Aug 20, 2026). You reach it through a /search/developer endpoint with result-type, repository, and documentation-source filters. Firecrawl also published DevDex, an open benchmark of 1,179 developer-search queries scored on Recall@10 and MRR@10, so the retrieval quality is measurable rather than asserted. What it is not: it does not store code, and it is not a general web-search endpoint. It is an artifact layer, and a general scraper does not have one.

ChatGPT plugin (Aug 5). Firecrawl is now an official ChatGPT plugin, installable inside ChatGPT in three steps. Once connected, it gives ChatGPT five actions in plain language: search the web with full page content returned, scrape any page or document, crawl a site, interact with dynamic pages (logins, forms, pagination), and monitor pages for changes with alerts that can post to Slack (firecrawl.dev/blog/firecrawl-official-chatgpt-plugin, Aug 5, 2026). The same toolkit works inside Codex, which is the part a developer cares about: search for how something works, then pull the exact docs to write against it, without leaving the conversation. This is distribution, not a new capability. If you already call the API, the plugin adds nothing to your stack, but it lowers the on-ramp for anyone who lives in ChatGPT.

AnyDoc and pdf-inspector (Aug 6). Firecrawl open-sourced the two Rust libraries that parse documents behind the scenes. pdf-inspector reads a PDF's internal structure to decide per page whether the content is native text or needs OCR, so text-based pages skip the GPU entirely; the repo carries 13k GitHub stars at time of writing (firecrawl.dev/blog/anydoc-and-pdf-inspector, Aug 6, 2026). AnyDoc handles 14 non-PDF formats (docx, xlsx, pptx, rtf, odt, epub, csv, and the rest) in one binary with no API key and no system dependencies, converting each to clean Markdown. Both already power /parse and /scrape, so if you use those endpoints you were using these libraries before they had names. The reason this is a pro and not a footnote: a RAG pipeline that dropped PDFs and Office files into Firecrawl now has an open-source, dependency-free fallback if it ever wants to run the parsing itself.

Research Index, life-sciences category (Aug 13). Flagged in the July review as an arXiv-focused niche, the Research Index expanded to cover 41M+ papers across drug-discovery, clinical-trial, and biology literature, refreshed daily, with the launch post citing 90% Recall@10 on its paper-retrieval eval (firecrawl.dev/blog/research-index-life-sciences-launch, Aug 13, 2026). The bigger change is the price: the whole Research Index, including the AI and ML literature it already carried, is now free to query through /search/research, the CLI, MCP, and SDKs. A life-sciences research agent gets citable papers and full-text-on-demand retrieval for zero credits. That widens the "when it earns its place" answer from July: it used to be a bet on the AI/ML vertical, and now it covers life sciences too, at no cost.

Replit Connector (Jul 22). Slightly older than the August batch but not in the July review: Firecrawl is an official Replit Connector, one click from the Integrations panel in a Replit Core, Pro, or Enterprise workspace (firecrawl.dev/blog/firecrawl-official-replit-connector, Jul 22, 2026). Once active, Replit Agent can call search, scrape, interact, monitor, crawl, and parse while building an app. Usage bills through Replit's key by default, or you paste your own key. New connectors carry a 10,000 free-credit offer. Same story as the ChatGPT plugin: distribution into a builder surface, not a new endpoint.

The credit math, re-checked for September

The dollar figures need a caveat this month. The pricing page renders its prices behind a JavaScript slider, and on the Sept 4 fetch the Hobby and Standard dollar amounts came back blank. What did render, and what I can confirm as current: every credit allowance, every concurrency cap, the per-endpoint credit costs, and the Scale tier at $599/mo billed yearly (firecrawl.dev/pricing, fetched 2026-09-04). The Hobby ($16) and Standard ($83) monthly-billed-yearly figures below are the last confirmed numbers from the July review (fetched 2026-07-14); the page structure and the yearly-savings badges are unchanged, but I could not re-render those two dollar amounts in this run, so treat them as July-confirmed, not September-fresh.

Credit consumption decides your tier, not page count. Scrape, Crawl, and Map cost 1 credit per page. Search is 2 credits per 10 results. Interact is 2 credits per browser minute. Monitor is 1 credit per page per check. The Agent endpoint gives 5 free runs a day, then switches to dynamic pricing. All of that re-rendered on the Sept 4 page (firecrawl.dev/pricing, fetched 2026-09-04).

Now the new hole. The Developer Index adds a /search/developer endpoint, and the pricing page does not break out its per-query credit cost. The page lists a generic Search at 2 credits per 10 results, but whether a developer-search query bills at that Search rate or on its own line is not documented. Until it is, /search/developer joins the endpoints you cannot budget in advance: the Agent endpoint's dynamic pricing, and the still-unpriced question, highlights, and Lockdown Mode formats from earlier releases. If a coding agent leans on the Developer Index for most of its retrieval, model that call volume separately and watch the meter for the first billing cycle rather than trusting a forecast.

Map the flat rates to a workload. A research agent that scrapes 200 pages every morning burns 200 credits a day, 6,000 a month, which sits just over the 5,000-credit Hobby ceiling, so Standard is the tier (firecrawl.dev/pricing, fetched 2026-09-04). That is the upgrade trigger: not a page count, but the moment steady-state burn crosses roughly 6,000 credits. Below it, Hobby covers you. Above it, Standard's 100,000 credits give you headroom to run monitors and the Developer Index without doing arithmetic every week.

Monitor is where the meter runs faster than people expect, because the check interval multiplies straight through the page count. Watch 50 competitor pages daily and that is 1,500 credits a month. Watch the same 50 hourly and it becomes 50 times 24, or 1,200 credits a day, roughly 36,000 a month, a 24x swing with no extra page added. That single change burns through the entire 5,000-credit Hobby allowance before the first week is out, and on Standard it eats 36,000 of your 100,000 credits. All arithmetic off the published 1-credit-per-page-per-check rate, so run it against your own cadence before you pick a plan. Frequency picks your tier. Page count barely gets a vote.

Who buys which plan

The plans did not change, so the picks are close to July, with the Developer Index shifting the reason a coding-agent team lands where it does.

That coding-agent row commits to Standard rather than Hobby because of the Developer Index. If your agent's whole job is pulling repo context, docs, and PR history, you want the concurrency and the credit headroom to run /search/developer alongside your scrapes, and you want to watch that first bill because the endpoint's credit cost is not published. For a solo builder still proving a pipeline, none of that applies yet, and Hobby or the free tier is the honest call.

The rating, updated

The rating moves to 4.6 from 4.5. The case is narrow and specific: the Developer Index gives a coding agent a semantic retrieval layer over real GitHub artifacts that a general scraper does not have, and it arrived at no change to the subscription price. Research Index access went fully free and grew into life sciences, so a feature that was a niche bet in July now costs nothing to try. The parsing stack is open-source and dependency-free. Distribution widened into ChatGPT and Replit. Same price, more capability for the exact reader this review serves.

The candid cons are the ceiling on that bump. That new /search/developer endpoint carries a credit cost the pricing page never breaks out, so it lands on the same pile as the Agent endpoint's dynamic pricing and the unpriced question, highlights, and Lockdown formats: four-plus features you cannot budget before you run them. Crawl credit math still bites, a 5,000-page crawl still clears Hobby in one run, and cheap-tier concurrency is still low. And this month the pricing page would not even render the Hobby and Standard dollar figures, so the strongest single fact in any review, the price, is July-confirmed rather than freshly read. That is why the move is 0.1 and not more.

For a coding-agent team, Firecrawl in September 2026 is the tool to buy and Standard is the tier: start on Firecrawl here and let the Developer Index carry the retrieval a raw scraper cannot. Rated 4.6 of 5.