TinyFish Review (2026): Pricing, Features & Honest Verdict

TinyFish is a web infrastructure API for AI agents: search, fetch, a hosted browser and a multi-step agent behind one key. We found the best free tier in the category and a concurrency ceiling you cannot buy past. Best for: teams building production web agents. Price: free Search and Fetch, Agent from $0.016 per step. Rating: 8/10.

What is TinyFish?

TinyFish sells the plumbing that AI agents need to touch the live web. Four products share one API key. Search returns structured results. Fetch turns any URL into clean content. Browser gives you a managed remote browser session. Agent runs multi-step automation across sites, including authenticated flows. There is also Mako, a web-native model the company positions for live workflows.

The company launched out of stealth in August 2025 with $47 million in Series A funding led by ICONIQ, joined by USVP, Mango Capital, MongoDB Ventures, ASG, and Sandberg Bernthal Venture Partners. It was founded in 2024 and runs out of Palo Alto with a team of roughly 25. The founders are Sudheesh Nair, formerly president of Nutanix, Shuhao Zhang, previously an engineering leader at Meta, and Keith Zhai, a former senior correspondent at The Wall Street Journal. That is an unusual mix, and it shows in the product: the pricing page reads like it was written by someone who has sold infrastructure before, and the benchmarks page reads like it was written by someone who expects to be fact-checked.

What Are TinyFish’s Key Features?

Search and Fetch that cost nothing at real rate limits

Search runs at 30 requests per minute and Fetch at 150 URLs per minute, both at $0.00. These are not trial allowances that expire. TinyFish states that both keep working at a $0 wallet balance and never draw from the wallet at all. For a lot of agent workloads, retrieval and page cleaning are the whole job, which means a real slice of users will never pay this company anything.

Agent runs with inference and proxies included

The Agent product bundles what a web run actually needs into one price. TinyFish includes the LLM inference (“no key to bring, no provider bill on top”), routes requests through residential IPs so sites serve the page a person would see, handles bot checks at the infrastructure layer, and keeps screenshots, logs and source URLs from every run. Compare that to tools where the subscription buys orchestration and you supply the model bill separately, and the $0.016 per step starts looking less like a rate and more like an all-in number.

A prepaid wallet instead of a plan

There are no plans, no seats, and no monthly minimum. You fund a wallet, Agent and Browser draw from it, and auto-reload refills it. New accounts start with $8 in wallet funds before you add a card at all. When the balance hits zero nothing breaks: Search and Fetch keep going, and an Agent run already in flight finishes rather than dying halfway.

MCP-native access from any client

All four products are reachable over MCP at a single endpoint, so Claude, Cursor, or any MCP-compatible client picks them up with one config change. SDKs and a CLI ship alongside, and a direct API is there for custom integrations. This matters more than it sounds: the difference between an agent tool you have to wrap and one your existing client already speaks is usually an afternoon.

How Much Does TinyFish Cost?

Search is $0.00 per request at 30 requests per minute. Fetch is $0.00 per URL at 150 URLs per minute. Agent is $0.016 per step with a 2 concurrent run limit. Browser is $0.002 per minute with a 5 concurrent session limit. The minimum wallet deposit is $10.00, and new accounts get $8 in wallet funds without a card. Enterprise is quoted, and adds ISO 27001, enterprise SSO, audit logs, VPC deployment, and a 99.99% uptime SLA.

Against the field, the free half is the standout. Browserbase includes 1,000 search calls and 1,000 fetch calls on its $20 per month Developer plan and then charges $7 per 1,000 search calls and $1 per 1,000 fetch calls. TinyFish charges nothing for either, ever. On the paid half the story flips to parity: Browser at $0.002 per minute works out to $0.12 per browser hour, which is exactly Browserbase’s Developer overage rate and above its Startup rate of $0.10 per hour. The difference is that there is no $20 or $99 subscription sitting underneath it.

What Is a “Step”, and Why Does It Matter?

Agent is the product you will spend the most on, and it is priced in a unit TinyFish never defines. We checked the pricing page and the public Agent documentation page and found no definition of a step anywhere. Browser is $0.002 per minute, which needs no explanation. Search and Fetch are free, so their unit does not matter. That leaves the single metered product that will dominate your bill as the one you cannot estimate in advance.

This is not a pricing scandal, but it is a planning problem. A five-page authenticated workflow might be five steps or fifty, and the difference between those two is the difference between $0.08 and $0.80 per run. At any real volume that is the whole budget conversation. Meter a representative workflow on the $8 welcome balance, which buys exactly 500 steps, and read your own numbers off the dashboard before you commit anything.

Can You Buy More Concurrency?

No, and this is the finding that should shape your decision. Accounts start at 2 concurrent Agent runs and 5 concurrent Browser sessions. The FAQ is admirably direct about what happens next: limits “live on your account rather than being bought with Wallet funds, so raising them is a conversation, not a purchase.”

So the “no plans, no minimums” pitch holds right up until you need throughput, at which point it becomes a sales call. Browserbase will sell you 25 concurrent browsers for $20 a month and 100 for $99, self-serve, today. TinyFish gives you 5 at any spend level. If your workload is a nightly batch of a few hundred pages, that ceiling never bites. If it is a user-facing feature where ten people might hit the same agent at once, you are talking to sales before you launch, and you should factor that timeline in now rather than discovering it during your load test.

Are the Benchmarks Trustworthy?

More than most. TinyFish publishes six comparisons and labels each one as either a public benchmark or its own eval, which is a standard almost nobody in this category meets. Four are public: Online-Mind2Web success rate at 89.9% against Gemini 2.5 Computer Use at 69.0%, OpenAI Operator at 61.3% and Claude Computer Use at 56.3%; WebVoyager task pass rate at 91.1% against BrowserUse at 88.3% and Smooth at 86.6%; SimpleQA first-result sufficiency at 49.2% against Tavily at 45.6%, Parallel at 43.6%, and Exa and Firecrawl both at 38.4%; and BrowseComp latency at 1.47 seconds against Exa at 1.59 and Firecrawl at 1.83.

The two in-house evals are marked as such: a Fetch quality eval at 93% usable pages against Tavily at 80% and Firecrawl at 62%, and a p50 search latency of 556ms against Exa at 811ms. Treat those two as vendor claims. The detail that earned the most credit here is SimpleQA at 49.2%: they lead the field on it, but it is still a number under half, and they published it rather than picking a friendlier metric.

Who is TinyFish Best For?

Use TinyFish if you are building agents that read the web at moderate concurrency and you want retrieval to be free. It is the strongest fit for teams whose agents search and fetch far more than they click, because that entire half of the workload costs nothing and the rate limits are high enough to be useful. It also suits anyone tired of stitching a proxy vendor, a captcha service and a model provider together, since the Agent price absorbs all three.

Skip TinyFish if you need more than 5 concurrent browser sessions without a sales conversation, or if you need a step-level cost estimate before you write a line of code. Skip it too if your compliance checklist demands SOC 2, since the published attestation is ISO 27001 and it sits on the Enterprise tier. Teams already deep in Browserbase’s stack with heavy concurrency needs will find the self-serve ceiling a downgrade even though the unit rates are competitive.

Best TinyFish Alternatives

Browserbase is the closest infrastructure comparison. Its free plan gives 3 concurrent browsers, 1 browser hour, 3 agent runs and 1,000 each of search and fetch calls. Developer is $20 per month for 25 concurrent browsers and 100 browser hours, then $0.12 per browser hour. Startup is $99 per month for 100 concurrent browsers and 500 hours at $0.10 per hour. You pay a subscription, and in exchange concurrency is something you can buy with a credit card.

Exa is the better comparison on search specifically. Its Starter tier is free with $20 in credits on signup and $10 in credits every month after, no payment method required, plus 5 queries per second and 3 agent concurrency. The recurring monthly credit is a real advantage over TinyFish’s one-time $8, though TinyFish beat Exa on both SimpleQA sufficiency and BrowseComp latency in the published numbers.

Firecrawl is the scraping-first option, with a free plan at 1,000 credits per month and 2 concurrent requests. Its paid tiers are set on a slider and geolocate their currency, so we are not quoting a figure we could not read cleanly. On TinyFish’s own published comparisons Firecrawl trailed on fetch quality and search latency, but it remains the more mature choice for plain large-scale crawling.

Final Verdict: Is TinyFish Worth It?

Yes, with your eyes open about throughput. The free half of this product is the most useful free tier in web infrastructure right now, and it is not a trial: unlimited-in-practice search and fetch at 30 and 150 per minute, working at a zero balance, from a company with $47 million in the bank and no obvious reason to yank it. The paid half is priced at parity with the market while bundling inference, residential proxies and anti-bot handling that competitors charge for separately. The benchmark page is labelled honestly, including a number under 50% that a less confident vendor would have buried.

The two things to settle before you build: get a real step count for your own workflow off the $8 welcome balance, because the pricing page will not tell you what a step is, and decide now whether 2 concurrent agent runs and 5 concurrent browser sessions will carry your launch. If both answers come back fine, this is the best value in the category. If your product needs to fan out, start the sales conversation on day one rather than the week you ship.