When people talk about “the” speedrun, they often mean a handful of games that defined what racing a video game to its limits looks like. In that short list, Super Mario Bros. for the Nintendo Entertainment System sits at the very top. Since its home console release in 1985, players have tried to clear it faster and cleaner, taking a simple thirty minute platformer and compressing it into a run that now lasts less than five minutes and leaves almost no input unused.
In the twenty first century that race has become something close to a laboratory. Community historians chart hundredths of seconds gained on obscure leaderboards. Programmers dissect the game’s code. Tool assisted runs show what perfect play looks like, and real time runners try to match it on actual hardware. By early 2026 the world record for the main any percent category stands at 4 minutes 54.415 seconds by the runner averge11, only a handful of frames away from the best time computers can produce.
This chronicle follows Super Mario Bros. as a speedrun game. It begins with a 1985 platformer designed to showcase a new console and ends with a global community fighting over individual frames, racing a timer that only checks their progress every twenty one frames, and treating a single level warp as one of the most studied pieces of platforming in the medium’s history.
A 1985 Platformer That Invited Races
Super Mario Bros. emerged from Nintendo’s R&D4 team as one of the NES’s showcase titles. Directed by Shigeru Miyamoto and Takashi Tezuka, it turned the simple “Jumpman” of Donkey Kong into Mario, a character who could sprint, skid, accelerate, and carry momentum through sprawling side scrolling stages instead of single screen boards. The game released on the Famicom in Japan in September 1985 and on the NES in North America later that year, with Nintendo citing an October 18, 1985 launch date in the United States.
For casual players the design emphasized exploration and novelty. Pipes hid secrets, invisible blocks offered power ups, and warp zones let a player leap from the early worlds straight into harder late game stages. For speedrunners, those same traits looked like an invitation. The game tracked completion with a visible timer. Every level had one clear goal, a flagpole or castle axe, that signaled success. The physics engine treated running speed and jump height consistently in every level, which meant that once a route was discovered, a player could repeat it from run to run.
Most importantly, Super Mario Bros. was everywhere. It shipped as a pack in game with many NES consoles, sold millions of copies, and became one of the most familiar cartridges in living rooms and arcades. Long before the term “speedrun” came into common use, children were already racing each other through World 1 1, bragging about who could “beat Mario” the fastest, and comparing times based on rough wristwatch timing or the dwindling in game clock.
High Scores, Twin Galaxies, and the Pre Internet Era
The earliest organized records for Super Mario Bros. came from the score and time culture that surrounded arcade games. Twin Galaxies, the American scoreboard organization that tracked high scores, eventually added NES categories and accepted videotaped runs of Super Mario Bros. in the late 1990s and early 2000s. At first the interest was not strictly about raw time. The organization used a different timing method, beginning from the moment a player pressed start rather than the frame Mario first came under player control, and rounded to whole seconds. It banned many of the glitches that would later define competitive any percent runs.
A retrospective compiled by Speed Demos Archive’s knowledge base calls this period “The Twin Galaxies Days.” It documents a sequence of records that fell steadily across the early 2000s. Rodrigo Lopes posted a 6:22 under that rule set, followed by Aaron Collins and Mike Morrow in the 5 minute range. By February 2004, Scott Kessler had pushed the time to 5:17, then to 5:14, and even lower with refined execution that still avoided techniques such as the wrong warp and wall jumps.
These runs already treated Super Mario Bros. as a race rather than a score chase, but they belonged to a world of mailed in tapes, private timing methods, and limited distribution. The modern speedrun community, which prizes transparent rules and global leaderboards, took shape in the mid 2000s around sites like Speed Demos Archive and, later, Speedrun.com. Super Mario Bros. came along for that transition.
AndrewG and the First Sub Five Minutes
When the community shifted away from Twin Galaxies’ restrictions, it embraced a different set of assumptions. Runners now timed their attempts from the first moment of player control, allowed glitches like the 4 2 wrong warp and flagpole trick, and obsessed over shaving off full seconds. In that environment the runner Andrew Gardikis, known as AndrewG, became the central figure.
The SDA knowledge base marks April 10, 2007 as the beginning of a new era with AndrewG’s 5:00.6 run, a historic time that incorporated warp pipes and a slower version of the wrong warp in World 8 4. Over the next few years he chipped away at that time, improving to 5:00.4 and then finally to 4:59.4 on December 24, 2010. A video titled “Super Mario Bros. Speed Run in 04:59!!!!!” captures that breakthrough and presents it as an almost impossible barrier falling in real time, with the run performed on an NES in one continuous segment.
AndrewG did not stop at simply breaking the five minute mark. He continued to improve his personal bests in tiny increments, reaching 4:58.89 and then refining that time across multiple records as he and other runners discovered new optimizations. These included hitting the top of flagpoles to minimize post level animations and despawning dangerous piranha plants in the crucial warp room of level 4 2.
By the early 2010s the broad outline of the any percent route was in place. A successful run would clear Worlds 1 1 and 1 2, take the warp to World 4, execute a precise set of maneuvers in 4 2 to reach a vine warp to World 8, and then survive the late game gauntlet. What remained were the tiny margins between frames.
The Frame Rule and the Science of Perfection
What makes Super Mario Bros. speedrunning unique is not just the quality of its route, but the way the game’s internal code interacts with the timer. The NES hardware updates the screen about sixty times per second. In Super Mario Bros., most of the counters that control timing are built on simple frame counting logic. For some of them, Nintendo’s programmers chose to update a “long” timer only once every twenty one frames.
As a result, almost every level transition only checks whether Mario has reached the goal at steps of twenty one frames, roughly 0.35 seconds. Community members call each of these windows a “frame rule.” If a runner finishes a level just after the check, they must wait until the next one. That means that two runs can differ by several visible frames at the end of a stage yet still arrive at the next level on the same schedule, with no difference in the official time.
Former world record holder Darbian popularized an analogy that compares the end of each level to a bus stop. Mario runs to the bus station at the end of the stage, but the bus only departs every twenty one frames. A runner who misses a bus by a single frame has to wait for the next one. Their only reward for an exceptionally clean level is catching an earlier bus.
For speedrunners this transforms the game into a specific kind of puzzle. It is not enough to play levels as fast as possible in isolation. The goal is to meet or beat the earliest possible frame rule at each stage while using movement patterns that feel consistent and repeatable. Tool assisted speedruns, which use emulators to test inputs frame by frame, revealed where those frame rules sit and which combinations of jumps, fireball shots, and enemy manipulations conserve the most time. TAS runs published on TASVideos by contributors like Pom, MrWint, HappyLee, and Mars608 showed that a perfect any percent run using warps could reach roughly 4:54.26 and that a full warpless run could be pushed under nineteen and a half minutes.
Those theoretical limits became targets. Human runners began to talk not just about seconds, but about how many frame rules away from the TAS they were, and which specific rooms or jumps still had time left on the table.
The Frame Wars, Darbian, and Kosmic
The mid 2010s brought a new generation of runners and a new level of precision. After AndrewG’s long run of records, a player known as i_o_l finally produced a 4:57.69 in 2014. That time, documented in the SDA knowledge base, represented the first run that both incorporated the bullet bill glitch and stayed close enough to TAS optimizations to reach a lower second boundary.
From there, times fell in tiny increments. The same SDA list calls the section that follows “The Frame Wars,” beginning with Darbian’s 4:57.62 in October 2015. Over the next few years he traded the record with Kosmic, lowering the mark to 4:56.878 and then 4:56.528. In February 2018 Kosmic pushed it slightly lower again to 4:56.462 with a run that tied or nearly tied the TAS in many stages while still losing a little time to lag and minor mistakes.
This period established many of the advanced techniques that later runners would rely on. Runners perfected backward acceleration jumps to reach full speed faster, learned to manipulate enemy positions for bullet bill glitches, and refined wall clips in the late castles. Summoning Salt’s documentary “The Super Mario Bros. Any% Odyssey” and similar videos helped popularize these innovations, explaining ideas like subpixels and frame rules to a wider audience and solidifying the game’s reputation as the archetypal speedrun.
By the time Darbian and Kosmic stepped away from the main record grind, the any percent world record sat within a couple of dozen frames of tool assisted perfection. Many community members believed that only a small number of players would have both the skill and the patience to push it further.
Niftski, averge11, and the 4:54 Era
That last phase of the race has been defined by two names. Starting in 2020, a runner known as Niftski began attempting any percent records and quickly took control of the leaderboard. In March of that year he set a 4:55.430 that improved on Kosmic’s best by a noticeable margin.
Over the next several years he lowered the record again and again, breaking the 4:55 barrier in 2021 with a 4:54.948 that Wired and other outlets compared to breaking a four minute mile in running. He refined his route with a technique known as “Lightning 4 2,” developed from TAS research by HappyLee and others, which manipulates platform cycles and enemy positions in level 4 2 to catch an earlier frame rule without sacrificing consistency.
By September 2023 Niftski had improved his time to 4:54.631, the first human run to match TAS frame rules in every level except the final stage 8 4. He followed that with a 4:54.565 in January 2025, leaving only a handful of frames between his time and the leading tool assisted record.
For the first half of the 2020s, no one else could take the crown for long. That changed in August 2025 when the runner averge11 set a 4:54.515 on MiSTer hardware that emulates the NES at a very accurate level. Games journalism sites noted that this new record came only about fifteen frames short of the 4:54.26 TAS benchmark and that it broke a four year stretch in which Niftski alone had held the any percent title.
The story did not end there. Between late 2025 and early 2026 the world record bounced between the two runners as they each tried to refine their execution in 8 4 and remove every possible lag frame. Speedrun.com’s leaderboard now lists averge11 back on top with a time of 4:54.415, set roughly a month before this chronicle was written, followed closely by Niftski at 4:54.448 and LeKukie at 4:54.831 on original NES hardware.
With these times in place, Super Mario Bros. is closer to its theoretical limit than almost any other major speedrun. The remaining gap is measured in a handful of frames scattered through 8 4. Any future record will likely require thousands of failed attempts to assemble one run where every jump, fireball, and elevator cycle lines up perfectly.
Warpless, Glitchless, and Minus World
Although any percent with warps receives the most attention, Super Mario Bros. supports a diverse set of categories, each exploring a different version of the game. Speedrun.com’s main board includes warpless, in which runners must clear every stage without warp zones; glitchless, which forbids tricks like wall clips and wrong warp; and minus world ending, a category built around reaching and escaping the game’s famous glitch level. It also tracks variants based on the Super Mario All Stars compilation and console differences.
The warpless category has its own long record progression, documented both on SDA and TASVideos. Early Twin Galaxies times were over twenty two minutes, but by 2013 runners like AndrewG and i_o_l had pushed the mark under nineteen minutes using fire flower routes and carefully planned power up grabs. Tool assisted warpless movies, like the 2018 submissions by MrWint and later improvements, served as guides for optimal power up usage and level pacing.
Glitchless and minus world categories, many of them defined and refined in the 2010s and 2020s, highlight different aspects of the game’s design. Glitchless showcases how far human skill can go while staying within the intended physics and platforming logic. Minus world runs, popularized in part by Summoning Salt’s work and various community videos, treat an old playground rumor as a structured challenge.
Collectively these categories turn Super Mario Bros. into more than one race. The same cartridge now hosts multiple distinct histories, each with their own pioneers, milestones, and theoretical limits.
Community, Documentation, and Cultural Impact
The reach of Mario’s first NES title has always extended well beyond the cartridge. In the context of speedrunning, it has provided both an entry point and a common language. Runners in many other games understand what a “frame rule” is because they learned it from Super Mario Bros. tutorials. The “bus stop” analogy has become part of the vocabulary of the broader community.
Historians and commentators have documented this evolution carefully. The Speed Demos Archive knowledge base provides a near primary narrative of early Twin Galaxies records and mid 2000s SDA runs. Speedrun.com serves as the live ledger of modern competition, listing thousands of submitted runs with dates, platforms, and verification notes. Video essays by Summoning Salt and blog posts like “The Super Mario Bros Any% Odyssey” explain concepts such as frame rules, subpixels, and warp room manipulations in accessible language, turning technical tricks into public stories.
Tool assisted communities have also left their mark. TASVideos hosts both the movies themselves and long forum threads in which authors explain why they chose certain strats, how they discovered new optimizations, and where they believe remaining time might still exist. For historians, those discussions read almost like lab notes from an experiment in pushing a fixed system to its theoretical breaking point.
All of this has made Super Mario Bros. more than a nostalgic curiosity. In 2025, as Nintendo and various media outlets celebrated the game’s fortieth anniversary, they highlighted not only its role in popularizing home consoles but also the way it continues to inspire competitive play decades later.