Bitfinex: crypto exchanges rating breakdown

Custody of funds: 3/10

Proof-of-reserves: does it exist, who attests it, does it include liabilities (a PoR without liabilities is close to meaningless), how often. Insurance fund: size, is the address public, and has it ever actually paid out. Hack history — and above all whether users were made whole. Licences by jurisdiction, and the enforcement record behind them. Every claim here is sourced to a primary document: a regulator’s register, a court filing, or the venue’s own published proof.

Scored 3 of 10 against a category median of 5, which places it 19th of 28 among centralised exchanges on this criterion. At a 40% weight that is 0.80 points below the median contribution of the weighted total. The best score in the category is 9, the worst 1.

A single blog post from November 2022 listing some wallet balances, and a promise to “continue to develop” a periodic proof. It never arrived. There is no proof-of-reserves page on the site today — the URLs simply 404. The in-house tooling it built to prove liabilities has not been touched on GitHub since 2018. Assets-only, four years old, and abandoned. Nothing that covers theft. Its insurance fund is defined in its own API documentation as “the balance available to the liquidation engine to absorb losses” — a derivatives backstop, nothing more. History matters here: after the 2016 theft of 119,755 BTC, Bitfinex socialised the loss with a 36% haircut applied to every customer balance, including accounts that were never touched, then issued tokens that were eventually redeemed. Users were made whole in the end — but only after everyone was forced to share the loss.

True cost of trading: 8/10

Measured by us, not read off a fee page. We walk the live order book and fill $1,000 / $10,000 / $100,000 of BTC/USDT, then add the venue’s published taker fee. That sum — fee + spread + realised slippage — is the effective cost. Separately: the markup on network withdrawal fees, and the hidden spread inside “convert” / “buy in one click” flows.

Scored 8 of 10 against a category median of 6, which places it 13th of 28 among centralised exchanges on this criterion. At a 16% weight that is 0.32 points above the median contribution of the weighted total. The best score in the category is 10, the worst 1.

Liquidity and execution: 7/10

Order-book depth within ±0.1% and ±0.5% of mid, measured from the live book. Uptime over 12 months from the status page. Behaviour under stress: were there outages on days the market moved more than 5%.

Scored 7 of 10 against a category median of 6, which places it 9th of 28 among centralised exchanges on this criterion. At a 11% weight that is 0.11 points above the median contribution of the weighted total. The best score in the category is 10, the worst 1.

Assets and networks: 3/10

Number of pairs, adjusted for quality: the share of listings delisted within a year. Which networks are actually supported for deposit and withdrawal — not merely “supports USDT”.

Scored 3 of 10 against a category median of 6.5, which places it 26th of 28 among centralised exchanges on this criterion. At a 9% weight that is 0.32 points below the median contribution of the weighted total. The best score in the category is 10, the worst 1.

~168 real spot pairs — a short, deliberate list

Fiat on/off ramp: 3/10

Which methods are actually available per country, their fees, and their limits per verification tier — read from the venue’s published terms and cited. Where a venue does not publish this, that opacity is itself the finding, and it scores accordingly.

Scored 3 of 10 against a category median of 5.5, which places it 19th of 28 among centralised exchanges on this criterion. At a 11% weight that is 0.28 points below the median contribution of the weighted total. The best score in the category is 10, the worst 1.

USD, EUR and GBP by wire, both ways — but gated behind full verification and a $10,000 minimum on manual wires. Not a retail rail.

KYC and access: 4/10

Verification tiers and their limits, what is genuinely usable without verification, and the country deny-list — from the venue’s published policy, cited. A venue that will not tell you your withdrawal ceiling before you sign up is marked down for it.

Scored 4 of 10 against a category median of 6, which places it 21st of 28 among centralised exchanges on this criterion. At a 8% weight that is 0.16 points below the median contribution of the weighted total. The best score in the category is 9, the worst 1.

Four tiers are published, but they gate features rather than state withdrawal ceilings; the numeric limits above the entry tier are not published. United States: Any “U.S. Person” is a Prohibited Person under its terms; retail is excluded outright, with a discretionary carve-out only for eligible contract participants. A 2021 New York Attorney General settlement ($18.5m) required it to stop trading with New Yorkers. European Union: None found in the ESMA register. No CASP authorisation found — but, unusually, its terms contain no EU restriction at all. Whether it cut off EEA users when the MiCA transition closed on 1 July 2026 could not be sourced, so we say so rather than assume.

Product, API and docs: 6/10

API documentation quality, whether rate limits are documented, whether a public status page with uptime history exists, and the API response latency we measured ourselves while pulling the order book.

Scored 6 of 10 against a category median of 5.5, which places it 10th of 28 among centralised exchanges on this criterion. At a 5% weight that is 0.03 points above the median contribution of the weighted total. The best score in the category is 9, the worst 4.

Documented API and rate limits, and a status page — though it shows an incident log with no uptime figures. Note its trading fees are now zero, which makes our measured cost for it unusually favourable.