PDF Password Length vs Crack Time — Complete Analysis

The exponential maths behind password length

The candidate space for a brute-force attack is charset_size^length. For lowercase (26): 26^6 = 308 million, 26^8 = 208 billion, 26^10 = 141 trillion. For mixed-case alphanumeric plus symbols (95 printable ASCII): 95^6 = 735 billion, 95^8 = 6.6 quadrillion, 95^10 = 59.9 quintillion.

That exponential growth means one extra character changes everything. A 6-character password may fall in minutes; an 8-character one of the same type may take days to centuries depending on the mode — the same two-character increase multiplies the search space by 95^2 = 9,025x.

In practice recovery uses three strategies that avoid searching the full brute-force space: dictionary attacks (testing real words and patterns), rule-based mutation (adding digits, capitalising, substituting), and mask attacks (constraining character positions by type). For human-chosen passwords these cut the effective search space by orders of magnitude.

Mode 10400 (40-bit RC4) — the 40-bit key space

Mode 10400 (40-bit RC4, PDF 1.1-1.3) is the weakest tier. The file key is only 40 bits: the password is hashed with MD5 and the digest truncated to five bytes, so the number of distinct keys is capped at 2^40. Two different attacks follow. Hashcat's mode 10400 searches candidate passwords, so length and character set do affect it. A direct key-space search over the 2^40 keys is length-independent and needs different tooling; in Hashcat it is exposed through the collider modes 10410 and 10420.

On a single RTX 5090, Hashcat's published benchmark puts mode 10400 at about 5.0 billion candidates per second, so a full 2^40 sweep is roughly four minutes of GPU time. What that yields is a working key — access to the document — not necessarily the original password string.

For the key-recovery path, password length is irrelevant: any padded password whose MD5 truncation matches yields the same key. A password search in the same mode is not length-independent. Either way, mode 10400 is the tier where access to the document is reliably recoverable.

Mode 10500 and 10600 (128-bit RC4 / AES-128, and Acrobat 9 AES-256)

Mode 10500 covers both 128-bit RC4 (V=2/R=3) and AES-128 (V=4/R=4): Hashcat parses both under the same mode because the password check is the same 50-iteration MD5 derivation. Measured on an RTX 5090 it runs at about 166 million candidates per second. That is a password search, not a key search, and a 500M-candidate dictionary-plus-rule pass takes about three seconds.

Mode 10600 is not AES-128: it is PDF 1.7 ExtensionLevel 3 (Acrobat 9, 2008) AES-256 at revision 5, whose key derivation is a single SHA-256. That makes it the fastest PDF password mode after 10400 — about 28,500 million candidates per second on an RTX 5090.

Brute-force mask times for mode 10600 at about 28,500 MH/s, by length and character set:

• 6-char lowercase (26^6 = 308M): about 0.01 seconds. Instant.
• 8-char lowercase (26^8 = 208B): about 7 seconds. Instant.
• 8-char alphanumeric (36^8 = 2.82T): about 99 seconds.
• 10-char lowercase (26^10 = 141T): about 1.4 hours. Not a barrier.
• 8-char full printable ASCII (95^8 = 6.6Q): about 2.7 days. Feasible on a cluster.

Mode 10700 (AES-256, revision 6) — the modern standard

Mode 10700 is AES-256 at revision 6 — the construction introduced with PDF 1.7 ExtensionLevel 8 (Acrobat X, 2010) and retained unchanged by PDF 2.0. Its derivation is Algorithm 2.B, which runs a minimum of 64 rounds, roughly 70 typical, each combining an AES-128-CBC pass over a repeated buffer with one SHA-256, SHA-384 or SHA-512 round. That makes it by far the slowest PDF mode per candidate: about 595,000 candidates per second on an RTX 5090, some four orders of magnitude below the revision-5 mode.

Dictionary-plus-rule attack (10M words, 50 rules, 500M candidates) at about 595 kH/s: about 14 minutes. Practical for human-chosen passwords.

Brute-force mask times for mode 10700 at about 595 kH/s, by length and character set:

• 6-char lowercase (308M): about 8.5 minutes.
• 7-char lowercase (8B): about 3.8 hours.
• 8-char lowercase (208B): about 4 days on one GPU; hours on a large cluster.
• 8-char alphanumeric (2.82T): about 55 days on one GPU; about 7 days on eight.
• 10-char lowercase (141T): about 7.5 years on one GPU. Out of reach on realistic clusters unless the password is human-chosen.

How clusters change the picture

The times above are for a single RTX 5090. A professional service with 8 GPUs reduces every time by roughly 7.5x, allowing for 8-GPU scaling overhead.

With 8 x RTX 5090 on mode 10700: 8-char lowercase drops from about 4 days to roughly 13 hours. 8-char alphanumeric drops from 55 days to about 7 days.

With a 32-GPU cluster (four 8-GPU systems) the rate is about 4x the 8-GPU figure: roughly 17.9 MH/s on mode 10700. 8-char lowercase drops to about 3.2 hours, and 8-char alphanumeric to about 1.8 days — costly but potentially worth it for a high-value document.

The practical ceiling: under mode 10700, 10-character lowercase is borderline — roughly three months on a 32-GPU cluster — and the infeasible boundary sits around 11-12 characters of a wider set (36^11 or 95^10). That is the line between recoverable and unrecoverable for mode 10700 PDFs.

How effective dictionaries and rules are

Brute-force times are worst-case. In practice most forgotten PDF passwords are NOT random. They are human-chosen: dictionary words, names, dates, patterns, keyboard walks. For those the dictionary and rule stage usually succeeds quickly, and pattern matters more than length.

Human passwords follow predictable shapes: lowercase base word, digits, an optional symbol. Public breach corpora run under the best66 ruleset (best64 up to Hashcat v6.2.6) capture most of these patterns. Success depends on whether the password follows common patterns, not just on its length.

A 12-character password may crack in minutes if it is 'iloveyou1234' — a known pattern plus a suffix rule. An 8-character random string like 'kX9#mP2q' would take centuries under mode 10700 despite being 4 characters shorter. Pattern matters more than length once dictionary attacks are in play.

Practical advice for password strength

For documents you want to be recoverable if forgotten: an 8-9 character password is the realistic ceiling, and the character set matters more than the length. Lowercase-only is a matter of hours to days on a professional cluster even under mode 10700; mixed-case letters plus digits (62 characters) is roughly five months on a 32-GPU cluster under mode 10700, and minutes under the Acrobat 9 revision-5 mode. Establish which revision your files use before relying on either.

For documents that must NOT be recoverable if forgotten: use a 12+ character random password from a password manager on an AES-128 or AES-256 file (modes 10500 to 10700). That is unrecoverable even with maximum GPU resources. If you forget it, the document is permanently lost — the appropriate level for truly sensitive material.

For legacy 40-bit RC4 PDFs (mode 10400): the key space is finite, so the document stays openable, but a password search in the same mode is not length-independent — a long random password still has to be searched. Either way the encryption provides no meaningful protection against a determined attacker.