Short answer: PHP 8.3 is security-only and reaches end of life on December 31, 2027. Plan one move to PHP 8.5, supported until December 31, 2029, using 8.4 as an interim stop if needed.
PHP 8.3 reaches end of life on December 31, 2027, per the php.net supported versions page, and from that day the branch gets no further security releases. Our advice is a single, deliberate move to PHP 8.5, which php.net backs with security support until December 31, 2029: fold the 8.4 and 8.5 migration notes into one compatibility pass, then shift workloads across on parallel FPM pools during 2027 rather than racing the calendar in December.
Source: php.net release calendar, verified on November 21, 2026.
php.net runs every branch on a fixed clock: twenty-four months of full maintenance, then twenty-four months where only critical security bugs get patched and point releases appear as needed. For 8.3, the switch to that second mode happened at the close of 2025. Routine fixes are already over, and the remaining safety net is removed when 2027 ends.
Why 8.5 rather than 8.4? Coverage per unit of effort. Finish in 2027 on 8.4 and you hold roughly twelve more months of patches before repeating the exercise; 8.4 also left full maintenance at the end of 2026. Land on 8.5 and you get about twenty-four months, with full maintenance still running throughout your rollout. The price is one extra set of release notes, which a long runway easily absorbs. If a critical library is not yet ready for 8.5, parking on 8.4 is acceptable, provided the onward move already has a date on the roadmap.
Once a branch is retired, php.net's guidance is short: upgrade soon, since unpatched holes are the expected outcome. The project's list of retired branches carries the same message for every version on it. From January 1, 2028, a fix that ships for 8.4, 8.5 or later will not be backported to 8.3.
Today 8.3 still receives critical patches, so there is no emergency. What waiting costs is room to maneuver: an upgrade begun early can pause, iterate and revert at will, while one begun in December 2027 has none of those luxuries.
There is a compliance thread too. In SOC 2, criterion CC7.1 deals with spotting and handling vulnerabilities, and a dated, owned upgrade ticket is simple to show an assessor; our page on SOC 2 for AI-assisted codebases goes deeper. Buyers' vendor-risk forms often want a list of runtimes along with confirmation that each is maintained upstream. During 2027, 8.3 still qualifies, and having its successor booked keeps the form easy to fill in.
One product can easily carry four or five separate PHP installs (web workers, queue consumers, cron, CI runners, laptops), and they drift apart over time. Use this checklist to build a register row for each.
Our free EOL deadline radar shortcuts the first pass: paste in composer.json, a Docker build file or a pipeline definition and you get back a report that lists PHP 8.3's December 31, 2027 end of life together with the other runtimes it spots, so the deadline is visible to whoever sets next year's roadmap.
An 8.3 to 8.5 move spans two releases, documented by php.net in 8.4's backward-incompatible changes, 8.4's deprecations, 8.5's backward-incompatible changes and 8.5's deprecations. These are the changes that most often decide how hard the move is.
From 8.5, Opcache is always compiled in and always loaded. The --enable-opcache and --disable-opcache build flags are gone, no opcache.so file is produced, and a zend_extension line that loads it now emits a warning. opcache.enable and opcache.enable_cli still work. Custom image builds and ini templates that install or load Opcache separately need a cleanup, or every process will start with a warning in its log.
8.5 deprecates driver-specific constants and methods on the PDO base class. PDO::MYSQL_ATTR_SSL_CA becomes Pdo\Mysql::ATTR_SSL_CA, PDO::pgsqlCopyFromArray() becomes Pdo\Pgsql::copyFromArray(), and so on for SQLite, ODBC, DBLIB and Firebird. The numeric values of PDO::FETCH_GROUP, FETCH_UNIQUE, FETCH_CLASSTYPE, FETCH_PROPS_LATE and FETCH_SERIALIZE changed, which matters only if you stored or compared raw integers. FETCH_CLASS constructor arguments now follow call_user_func_array rules, so string keys act as named arguments. From the 8.4 side, PDO_MYSQL's autocommit, emulate-prepares and direct-query attributes are now booleans. The "uri:" DSN scheme is deprecated for security reasons.
8.5 warns when a float, or a string that looks like one, is cast to int and cannot be represented, and when NAN is cast to another type. Destructuring a non-array value other than null with [] or list() now warns. Using null as an array key, or passing it to array_key_exists(), is deprecated; use an empty string. These warnings tend to surface in import jobs and report generators that handle messy input, so run those against real data in staging.
The backtick shell operator, the long cast names, case statements closed with a semicolon and returning null from __debugInfo() are all deprecated in 8.5. __sleep() and __wakeup() are soft-deprecated in favor of __serialize() and __unserialize(). Incrementing non-numeric strings is now deprecated outright, so str_increment() is the replacement. Most of this is search-and-replace, which makes it a good first batch of pull requests.
curl_close(), curl_share_close(), finfo_close(), imagedestroy() and xml_parser_free() are deprecated because their handles are objects that PHP frees on its own. $http_response_header gives way to http_get_last_response_headers(). Calling chr() outside 0 to 255, or ord() on a string longer than one byte, is deprecated too, since php.net treats both as signs of a bug.
8.5 removes the disable_classes ini setting outright, because it broke engine assumptions. If a security baseline relied on it, that control silently disappears, so document the replacement before cutover. Deriving argc and argv from the query string for non-CLI SAPIs is deprecated; php.net suggests register_argc_argv=0. Teams that tuned session ID length or bits per character should know that 8.4 already deprecated those knobs.
Jumping two releases means the intermediate notes apply even though no server ever runs 8.4. That step moved four extensions to PECL, deprecated several long-tolerated signatures, turned some handles into objects and raised library minimums; our PHP 8.2 guide walks through that layer item by item, and php.net's list of further 8.4 changes fills in the remainder. Work through it first, since its fixes are prerequisites for a clean 8.5 test run.
A long runway favors a two-track approach. Track one changes code so it runs on 8.3 and 8.5 alike; track two swaps the interpreter later. Because nearly every deprecation fix is backward compatible with 8.3, track one can merge continuously, months before any production host changes.
Complete the exposure table per repository and environment, with one accountable name per row. Then measure 8.3 as it runs today: warnings per hour, p95 response time on the heaviest routes, memory per FPM child, jobs processed per minute and CPU spent on authentication. Flag early any service with an IMAP or Oracle dependency, since it defines the longest path.
Introduce an advisory 8.5 job in CI. A php -l sweep with the 8.5 binary catches syntax breaks in seconds; the full suite then runs with error_reporting at E_ALL, writing deprecations to their own artifact so reviewers watch the number shrink with each merge. Promote the job to required once it is clean. On the new images, php --ini=diff lists every setting that departs from built-in defaults, which makes old-versus-new configuration reviews quick.
Run 8.5 containers or pools beside their 8.3 counterparts on an identical code release. Queue consumers and back-office tools go first because problems there are visible and inexpensive. Next, the load balancer moves a growing percentage of public requests, with checkout and sign-in held until last, and each ring is measured against its 8.3 numbers before it widens.
A service graduates to 100 percent on 8.5 once a routine release has passed without incident. The paired 8.3 image stays tagged so reverting means pointing traffic back; no schema or data migration is involved, which keeps that path cheap. Target completion comfortably ahead of December 2027, leaving time to absorb a surprise while 8.3 is still patched.
Move composer.json to require 8.5 (constraint and pinned platform alike), regenerate the lockfile, and strip out every 8.3 container, OS package and pipeline job. Remove stale zend_extension lines that load Opcache. Finally, log 8.5's December 31, 2029 retirement date so the following upgrade begins as a scheduled item rather than a surprise.
This kind of program is a steady stream of small merges followed by one careful rollout. Staff it with a lead who carries the deadline, developers fluent in modern PHP who chip away at deprecations in each codebase, and someone strong on infrastructure for containers, pools, pipelines and traffic weights. Borrow a security engineer briefly if hardening relied on disable_classes or custom session settings.
Many teams begin with a pod of 3 to 5 senior engineers (a tech lead plus engineers), and a full pod is about seven. Ryz engineers on your team start from $7,000 per month (our published range is $7,000–$15,000 per engineer per month, mostly mid-level to senior), quoted per team. Ryz engineers work on your team, reporting to your leads. Talk to us about PHP developers, or about DevOps consulting to handle containers and pipelines, and scope your team.
Another model suits you better if the job is a few hours of freelance version bumping, or you need engineers working European or Asian hours. Teams running PHP 5 or 7 code face a longer road, covered in the modernization runbook for older PHP.
For now, yes in the narrow sense: critical security flaws are fixed until December 31, 2027, though everyday bug fixing ended in 2025. After that, upstream releases cease, so any flaw found later in the interpreter or a bundled extension stays open on 8.3 permanently. From then on, your own controls are the only defense.
Aim for 8.5. Its security coverage lasts until December 31, 2029, with active maintenance through 2027, so it stays fully maintained during your rollout. Two sets of migration notes apply, a manageable load given a year of notice. An interim stop on 8.4, patched until December 31, 2028, works if one dependency lags, as long as the follow-on move is scheduled.
Duration follows a handful of variables: how many separate interpreters you operate, whether an Oracle or mail extension must be rebuilt from PECL, how often backticks, old cast names and base-class PDO constants appear, test depth around batch and database code, and the number of approvers for traffic changes. Once the inventory exists, you can estimate from evidence.
Not from upstream, as far as the official pages show: the PHP project's support pages describe no paid extension for retired branches. Whoever supplies your PHP binaries, whether a Linux distribution or a managed platform, sets its own patch policy, so confirm with them what 8.3 coverage, if any, continues into 2028.
Pricing begins at $7,000 a month for each Ryz engineer and tops out around $15,000 for most roles, with the bulk of people mid-level or senior. Multiply headcount by months by rate and you have the budget. Before kickoff you receive the plan, the number and the people by name.
Questions we didn't answer? Email info@ryzlabs.com.
Thanks — your message has been sent. We’ll get back to you soon.
Something went wrong while sending your message. Please try again or email info@ryzlabs.com.