The Super Mario Bros Any% Odyssey

Super Mario Bros. is probably one of the most recognizable games ever. Since its release in 1985, the game has exceeded 58 million in copies sold as one of the best-selling video games of all time, and until the late 2000s, was THE best-selling video game of all time.

With a game having an iconic status in video game history, it was natural for many gamers to gravitate towards in search of how to optimize the gameplay. While players of other iconic games of that era such as Donkey Kong and Pac-Man focused on achieving high scores, Super Mario Bros. players focused on beating the game in the fastest time possible. This aspect of beating a video game as quickly as possible is a feat that is now known as speedrunning.

It goes without saying that the Super Mario Bros. speedrunning has one of the largest audiences due to its status in video game history, and has remained popular as a game to both speedrun and to watch. Whenever a monumental world record was set, the public sphere would catch a glimpse and be amazed with the skill of the speedrunner, with articles flooding the internet remarking the incredible achievements. The latest world record was set by speedrunner Niftski, who was the first to beat the game in under 4 minutes and 55 seconds (i.e. the first-ever 4:54), which was once thought to be humanly-impossible. With history set again, content creators on YouTube took a moment make their remarks, including Karl Jobst…

Again, I must re-iterate that Niftski’s run is an astounding achievement in not just Super Mario Bros. speedrunning history, but speedrunning history as a whole. I mentioned before that a time under 4 minutes and 55 seconds was thought to be beyond human reach, and I’d like to take a deep dive into how we got today now with the first-ever 4:54 having been set.

Important Need-to-knows

Before we begin our retrospect, I need to introduce some terminology to make the reading a little easier.

1) Glitches

The term “glitch” is a bit of a buzzword nowadays, so it’s best set the record straight on what it actually means: a glitch is merely a malfunction of a game’s code. When a glitch is performed, the game will behave differently than originally intended as it was programmed, and this will usually show visually depending on the glitch. That being said, many glitches in video games are difficult to execute properly.

2) Subpixels

A “subpixel” is a unit defined to be smaller than a pixel, with pixels being the individual squares that render the game visually. While these units are not seen visually, they are used to calculate the position of objects at a finer level than a pixel, and are often used to detect collision between objects. In Super Mario Bros., a pixel is divided into 256 sub-pixels.

3) Framerules

This is probably the most important term with regards to speedrunning Super Mario Bros.. Upon finishing a level (with two exceptions), the game will load the next level every 21 frames, and these 21 frames are referred to as a framerule, which is short for the “21-frame rule”. As Super Mario Bros. runs at about 60 frames per second, a framerule is about 0.35 seconds.

The most famous analogy made to understanding framerules came from former world record holder darbian. In the video above, he likens a framerule to being like a bus: for every framerule, at the end of the level, a bus will take Mario to the next level, and for every framerule, there will be a bus.

So imagine Players A and B playing a normal level, starting at the same time, and Player A plays perfectly. As long as Player B plays within a certain window of frames, both Player A and Player B will be on the same framerule and they will start the next level at the same time. Thus, with many playing the game, time is usually saved in increments of 21 frames, or about 0.35 seconds.

The two exceptions come from lag frames, in which the game lags (usually from an overabundance of objects), and the final level. The latter exception is particularly crucial to understand, as timing for a “run” ends the moment that Mario hits the axe at the end of the level. Thus, every frame counts in the final level.

With that out of the way, it’s time to go back in time.

Chapter 1: Speedrunning Infancy

The fastest way to beat Super Mario Bros. became the basis for speedrunning the game: Beat 1-1 normally, take the Warp Zone at the end of 1-2 to go to World 4, beat 4-1 normally, take the Warp Zone in 4-2 to go to World 8, then beat the rest of World 8 normally. As the levels in World 8 are particularly difficult, the run is considered serious starting from 8-1.

Scott Kessler and Trevor Seguin are considered the first true speedrunners of the game: Kessler beat the game in 5:10 on April 23, 2004, then Seguin beat it in 5:07 on October 9, 2004. Seguin then beat his record just under a month later on November 7, 2004 with a final time of 5:06. Kessler would retake the record with a time of 5:05 on September 14, 2006.

With each record, general small mistakes were cleaned up, and it seemed that with Kessler’s run, the record could not be taken down further. In addition, Twin Galaxies, which was the main organization that recorded achievements such as speedrunning records and high scores at the time, banned the execution of glitches.

Chapter 2: Breaking the 5-minute barrier

This would begin the separation from the glitchless any% run (the fastest route without glitches) and the general any%. With Twin Galaxies banning glitches, one speedrunner sought to build together a run that could separate itself from Kessler’s. Enter Andrew Gardikis, better known as AndrewG.

AndrewG had previously tied Seguin’s 5:06 prior to Kessler’s 5:05, and he began searching on his own for months on new strategies to break Kessler’s record, and he fixated on two aspects of the final level: a wall-jump off a pipe in the second room, which is a frame-perfect and pixel-perfect glitch, along with a wrong warp in the third room. How a wrong warp works is that the game can only have one “warp zone” area loaded at a time, meaning that if players scroll far enough in an area, they will be taken to a different screen. For the wrong warp in the third area, players are instead taken to the water section instead of being looped to the beginning of the third area.

However, AndrewG wanted more, and found that a wrong warp also existed in 4-2. Originally, players would have to hit the vine block and do a rather precise jump off a hidden coin block to grab onto the vine to climb into the warp zone. However, if Mario is somehow manipulated to be far enough right on the screen, by taking a pipe that normally drops Mario in a coin cache, Mario will instead be taken to the warp zone that will take him to World 8 as the coin cache will not be loaded yet. In addition, the vine animation is skipped entirely, which saves a large chunk of time.

In order to get Mario further right on the screen, the player has to jump backwards into objects in a certain manner, a method referred to as “bumping“. The interaction that occurs from the bump is that Mario will move horizontally, but the screen does not, which allows Mario to be further right on the screen than he is normally allowed to be. Three bumps became the most popular way to “bump”, and these would later be optimized, i.e. where to “bump”.

One thing to note as I present videos of notable runs is that with each run in 4-2 from this point forward, you may notice that the vine block is still hit. This is done to de-spawn the piranha that occupies the pipe used to execute the wrong-warp.

With his new strategies, AndrewG shattered Kessler’s record and ushered in the new age of Super Mario Bros. speedrunning. This then begged the question of whether or not the game could be beaten in under five minutes. After another few months of grinding and cleaning some slight imperfections, AndrewG broke the five minute barrier by setting the first 4:59.

The next breakthrough came with realizing on where to even hit the flagpole at the end of most levels. When hitting the top of the flagpole, Mario will barely miss the block at the bottom as he hops off. This is because by the game’s programming, Mario actually accelerates faster in the direction opposite of which he is facing. As a result, when missing the block at the bottom of the flagpole, Mario will accelerate faster towards the castle, and in many cases, this is barely enough to save another framerule. With a further optimized 4-2, AndrewG would take the record down to a 4:58…

Chapter 3: Breaking 4:57

AndrewG would further improve his record and streamed them on Twitch.tv, taking it down to a 4:58.56. He now wondered if 4:57 was possible and began the grind again. Two eventual strategies came to mind: saving another framerule in 4-2, or go for the extremely unlikely Bullet Bill Glitch. Allow me to explain what these entailed…

As mentioned before, 4-2’s optimal strategy up through AndrewG’s 4:58.56 involved an optimized method of performing three bumps in specific spots. Each bump pushes Mario to the right by between seven to ten pixels, and how much he would be pushed is somewhat random. In order for the wrong warp to work, Mario needs to be pushed at least 20 pixels to the right. If a method was found to get two bumps and get 20 pixels, you could potentially save a framerule. As it turns out, there was such a way to do so, and AndrewG would find it: bump against the lowest coin block on the left past the small tunnel of blocks, then bump against the elevator platform. This method would become known as “Fast 4-2” for its ability to save a framerule over the three-bump method. However, this method was also incredibly difficult to execute, with the chances of getting the 20 pixels needed to be about 25%. Additionally, players needed to go down the pipe at the first possible pixel while not scrolling the screen forward further, as failure results in arriving at the coin cache and killing the run.

The Bullet Bill Glitch (also known in short as “BBG“) is probably one of the most iconic glitches in Super Mario Bros. speedrunning, and it involves using a Bullet Bill at the end of 8-2 (hence the name). By jumping off the Bullet Bill low enough to the ground, it is possible for Mario to clip into the block at the bottom of the flagpole to end the level while skipping the animation of the flag coming down. Additionally, because Mario is low enough, Mario is instead pushed in front of the block at the bottom of the flagpole and the time remaining immediately starts getting counted for the score. Normally, Mario will go into the castle and seemingly disappear. He is actually sent to the background layer, and he’s actually in front of a block at the castle entryway, and this is when the time remaining starts getting counted to the score. As a result, executing the BBG successfully saves two framerules as opposed to one. However, it was lucky getting the blaster that shoots the Bullet Bill to even shoot fast enough, and getting a low enough jump is also difficult: the game checks for hit collision every other frame, so Mario will jump off of enemies differently depending on the frame. In order for a low enough jump to execute BBG correctly, the game has to detect Mario’s collision with the Bullet Bill on the second frame, and this 50-50 chance became known as the “coin-toss” of the speedrun.

As the BBG seemed random, AndrewG ruled it out in getting a 4:57 and fixated on the still incredibly difficult Fast 4-2 strategy which still gave him some level of control. Despite coming close on multiple occasions, however, AndrewG could only get his time down to a 4:58.09, which he achieved on March 25, 2014.

On June 25, 2014, speedrunner Saradoc/Blubbler beat AndrewG’s time and set the first ever 4:57 with a 4:57.693, and it was done on an emulator, which is allowed by the rules on speedrun.com (the current archive of all speedrunning records).

The most important aspect of this speedrun is that Saradoc/Blubbler did NOT go for Fast 4-2, instead opting for the BBG, a first for a world record time. One thing to remember is that because of BBG saves two framerules, Saradoc/Blubbler was going to be one framerule ahead of AndrewG, hence the large difference in time.

So how did the Bullet Bill even shoot? Well, as it turns out, framerules determine the enemy patterns, so if players booted the game and played on the same framerules, the enemy patterns will largely be the same. This would be the breakthrough for speedrunners, including AndrewG, to start incorporating BBG in runs.

As for AndrewG, he would eventually get his own 4:57, although this was after more records were set, as a new speedrunner was on the loose…

Chapter 4: The rise of darbian and 4:56

On May 14, 2015, then-fairly unknown speedrunner darbian set a 4:57 of his own with a 4:57.793, which was also the first done on console. Two days later, he beat his own Personal Best (or PB) again with a 4:57.709. For each of his attempts, darbian opted not to go for Fast 4-2 while going for BBG, and eventually broke Saradoc’s/Blubbler’s record by four frames after over 6,000 attempts with a 4:57.627 (video here).

At this point, in order to shave more time off the record, runners needed to incorporate BOTH Fast 4-2 and BBG, and the race was on between darbian, AndrewG, and another runner, Kosmicd12 (better known as Kosmic), who was AndrewG’s closest competitor prior to the first 4:57. Again, since both glitches were extremely difficult to pull off, it seemed unimaginable to have both in the same run.

However, in December 2015, AndrewG became the first to get them both in a run. Even crazier, he managed both the pipe jump and wrong warp in 8-4, but a slow water section prevented getting a world record. Despite this, it was still a personal best, as he now had his own 4:57.

darbian, however, would be the first to put together a world record time that had both in a run with a 4:57.427, and would further shave it down twice with a 4:57.260 and then a 4:57.244 (the last time being a difference of just one frame and doing so on accident after claiming he was done). Without any other notable innovations, though, it seemed that once again, the category was close to death.

However, speedrunner mav6771 broke through by discovering a way to execute a glitch known as the Flagpole Glitch (FPG): by performing a series of frame-perfect inputs, it was possible to clip into the block at the bottom of a flagpole in a manner similar to the BBG, except Mario isn’t low enough and still runs to the castle (although the flag animation is still skipped). This does, however, save 15 frames, which is almost always enough to save a framerule. The glitch was thought to be exclusive to Tool-Assisted Speedruns (better known as TAS), where programmers effectively program a series of inputs on specific frames to execute into the game. With the discovery of how a human could pull off the glitch, darbian became the first to successfully do so in a run (clip here). This meant that a 4:56, which was previously thought to be impossible for humans, was within reach. However, the chances of getting FPG was still too low, along with having to do Fast 4-2 and BBG, meaning it was placed on hold again.

This was when Sockfolder, who had been involved in TAS’s for Super Mario Bros., stepped in, and found a series of inputs that could gain runners favorable subpixels for FPG to work, which were normally out of a runner’s control. FPG was found to be a viable way to save time in 1-1 as had been done previously with mav6771’s strategy, but also in 4-1 and 8-3. 8-1 is a bit of a unique story in that FPG normally doesn’t save enough time, while 1-2, 4-2, and 8-4 lack flagpoles, and 8-2 already has the BBG which saves even more time. The setup for 8-3 was also found to be incredibly difficult, so it was also ruled out. Both darbian and Kosmic opted to incorporate FPG while reverting to the original 4-2 strategy of three bumps. Kosmic would be the first find success by tying darbian’s world record before setting his own with a time of 4:57.194 (video here).

However, despite Kosmic successfully pulling off BBG, the blaster shot later than desired, meaning he lost a framerule, meaning there was still a sizable chunk of time to be saved. Three days after Kosmic set the world record, darbian became the first to achieve a 4:56 with a 4:56.878, in what has since become the most-viewed speedrun for Super Mario Bros. on YouTube with over 12 million views.

Chapter 5: The age of clipping and 4:55

As I mentioned before, the Flagpole Glitch in 8-3 was considered the most difficult of the three to execute at the time of darbian’s 4:56.878. However, if you haven’t noticed already, speedrunners will do anything to save time in a way that is humanly possible, and darbian began to implement it in his speedruns. He’d eventually successfully pull of the 8-3 FPG in a run and save a framerule in another world record time about a year later with a time of 4:56.528.

However, you might now be wondering… would anyone opt to even try all three flagpole glitches, BBG, and the infamous Fast 4-2 strategy? As you might have observed, when going for 4:56, both darbian and Kosmic had abandoned Fast 4-2. Despite this, in the interim between darbian’s 4:56.878 and 4:56.528, AndrewG had begun experimenting with a TAS strategy of clipping into the blocks before the short tunnel in 4-2 to be pushed forward that dated to 2004. Previously when attempting it, he was unsuccessful, but it was now 2017 and his abilities in the game had improved significantly since, and he managed to pull it off in practice. However, his success began to gradually decline, and it seemed like 4-2 still needed a reliable strategy.

Then, a source with the one of the oddest usernames ever, xx_420_blazit_xx, discovered a strategy similar to the TAS strategy by clipping into the blocks. The main difference was that blazit jumped into the blocks that were a row higher, which proved to be more reliable (his original post can be found here): the subpixels needed to clip into the blocks were the same as the ones that players start with in the level, and it came down to execution of the commands to effectively “pull” Mario into the blocks and be pushed forward on the screen. However, the payoff was huge: if executed quick enough, it would allow players to save the Fast 4-2 framerule. More on this can be found in a video at the end of the blog.

Anyway, back to darbian’s world record. You might notice that after setting his new time, darbian quips, “the tie has been broken”. This was because Kosmic had managed to get to a 4:56.994 (video here) before tying darbian’s previous record with a 4:56.878 (video here, audio is out of sync). However, the blazit clip was found a couple months before darbian set his 4:56.528, and it gave the speedrunners a new strategy to tackle 4-2. Soon enough, Kosmic began to implement it and managed to set a 4:56.545 (video here). In his run, Kosmic used the blazit clip but not the 8-3 FPG, meaning he was tied with darbian entering 8-4. Unfortunately, he played 8-4 one frame slower, meaning he wound up finishing one frame behind darbian’s world record. However, that would change in about a week.

Kosmic would eventually beat darbian’s world record with a 4:56.462. Notably, though, he did end up getting one lag frame while executing BBG, although he still played 8-4 five frames faster, netting a save of four frames.

What was also notable about this run was that it spawned an explanation video on the aspects of the run by fellow speedrunner Bismuth, who is also a moderator for Super Mario Bros. on speedrun.com, which brought an even larger audience to the Super Mario Bros. speedrunning scene. The video will be at the end of this blog.

A few months later, speedrunner somewes, who was then the third place runner, managed to beat Kosmic’s speedrun with a time of 4:56.245. One of the more comical moments of his run was his accidental kill of a Bullet Bill at the start of 8-3, which would change the Hammer Bros. patterns in the level (this is an instance of the “framerules dictate enemy patterns” notion breaking), which brought a substantial amount of stress. In that same level, however, he managed to execute the 8-3 FPG, which by that point had a setup that was significantly easier to pull off.

After setting the new world record, somewes had a new goal: become the first to set a 4:55, which had long been considered a pipe-dream, pun intended. However, there was a strategy out already to save an additional framerule to match the TAS: the 1-2 pipe clip (now see why I said “pun intended”? :3). By playing through 1-2 without slowing down, a difficult feat in and of itself, and then performing a pixel-perfect, frame-perfect jump with specific inputs to clip into the pipe at the end of the level, then walking forward into the warp zone, you will save the final framerule. This must sound crazy (as if speedrunning for 4:57, let alone 4:56, is crazy enough), and unsurprisingly, this is considered the single hardest part of going for 4:55. In fact, no one was capable of performing the clip until 2017 when AndrewG was the first to do so in practice (link here), around the same time when he was experimenting with clipping in 4-2.

Following his new world record, somewes made 6022 attempts at the first 4:55. Of them, only 795 managed to successfully clip into the pipe in 1-2, and only 442 of those managed to complete the level on 4:55 pace, a success rate of about 7.3%. To further show how difficult attaining 4:55 is, only two of his attempts managed to get to 8-4, and both of those runs failed at the pipe jump.

Meanwhile, Kosmic had been making attempts of his own to beat somewes’s record, and on September 24, 2018, he managed to tie it (video here). Afterward, he decided to do some 4:55 attempts for fun. His first eight runs were rather uneventful as he failed to get past 1-1. However, on his ninth run, he managed to successfully clip into the pipe in 1-2 and match the TAS framerule to be a framerule ahead of his new record. However, he then managed miraculously pulled off the remainder of the run, including BBG without the lag frame and the 8-3 FPG, for a final time of 4:55.913, the first ever 4:55. Kosmic’s shock throughout the run shows just how insane his run was (such as after BBG and even more so after 8-3). This would also prompt Bismuth to get to work on a video on what was perceived to be the final second barrier ever.

However, you might have also noted that Kosmic quipped, “Whatever, it’s slow, get 4:55” as he went into the water section in 8-4. Indeed, while his 8-4 was fast enough to get 4:55, it still was not particularly optimal. This was where somewes stepped in and redeemed his failures. As Bismuth was editing his video on Kosmic’s historic run, somewes snagged the record by playing 8-4 faster with a final time of 4:55.796.

About 10 months later, speedrunner Taven Webb would beat somewes’s record with a time of 4:55.746. While the run drew scrutiny given that it was not recorded via capture card, it was confirmed to be a legitimate speedrun. Indeed, Taven had previously performed multiple speedruns and posted them to his YouTube account, including one instance of accidentally performing the floor clip in 4-2. Additionally, he would also invest in better equipment to both stream and record his runs, and also managed to achieve the best 8-4 in a full game run (video here). While he hasn’t been as active in the last several months, it goes without saying that Taven’s successes are legitimate and should be regarded as one of the best speedrunners of Super Mario Bros..

Kosmic would eventually improve on his miracle run by finishing 8-4 faster and recording a 4:55.646, in what he would dub as his final personal best, stating that he had achieved all he wanted in a single run. Additionally, he made mention that while the time was improvable, he wanted no part in it due to the nature of the feats required to do so, and that it was “history for someone else to write”.

Chapter 6: The next discovery and the final second barrier

You might be wondering why Kosmic would make such a remark. A legend of Super Mario Bros. speedrunning, saying he wants to defer the title of “world record holder”? After all, this was also the first person to break the 19-minute barrier in the Warpless category, which further cemented his legend.

Indeed, had you watched Bismuth’s second “Speedrunning Explained” video on Kosmic’s 4:55.913 and somewes’s 4:55.796, you might have also considered the idea of beating Kosmic’s time preposterous, as Bismuth considered the human limit to be around Kosmic’s time.

Furthermore, Bismuth laid out two possible routes to getting a 4:54.998 (the slowest possible 4:54) or under:

Route 1: 1-1 FPG, 1-2 pipe clip, 4-1 FPG, “TAS” 4-2, 8-1 FPG, BBG, 8-3 FPG, save 3 frames in 8-4
Route 2: 1-1 FPG, 1-2 pipe clip, 4-1 FPG, blazit clip in 4-2, 8-1 FPG, BBG, 8-3 FPG, save 24 frames in 8-4

The second he ruled out as saving 24 frames in 8-4 at the time of the video’s posting meant playing just one frame behind the TAS, which is considered impossible given the constraints and conditions of a full run. However, he also ruled out the first as no one had matched the TAS 4-2 framerule in practice even with new “humanly-possible” strategies, and that executing the 8-1 FPG while saving a framerule (which I will explain later) was chalked to be a 1 in 1000 chance.

Lastly, the RTA (Real-Time Attack, which are speedruns accomplished by humans) 8-2 framerule at the time was one framerule behind the TAS framerule, and was dubbed “the least likely to ever occur” due to the level of precision needed to save it. However, breakthroughs were found to make it RTA-viable, and Taven was one of the first to do so in practice (attempt here).

However, Taven’s method was determined to still be somewhat random, and it was back to the drawing board… but not for long, as Kriller found more consistent setups to allow for a faster BBG to save a framerule. Eventually, a new wave of speedrunners came in, one of whom was Niftski, who managed to use the new strategy in a run to beat Kosmic’s time with a time of 4:55.430. However, you may notice that this was not at least a framerule faster, meaning that the 8-4 time was slower.

One of the other newer speedrunners who began to implement Kriller’s 8-2 framerule setups was Miniland, and he would eventually set a new record of 4:55.314 earlier this year (as of this blog’s original posting), then beat it further with a 4:55.230 about a month later. At this point, 4:54 seemed imminent. What was thought to be impossible suddenly looked within reach. To show just how much amazing this realization was, darbian, who had been over three years removed since his PB, had previously left a long comment on Niftski’s video of his 4:55.430:

“It seems so silly now to look back at how it was pondered whether BBG could ever happen in a run, or whether anyone would ever achieve fast 4-2 AND BBG in a single run, or whether anyone would achieve FPG (2x, 3x), or whether anyone would achieve 1-2g, or whether anyone would put all of that together, and now whether anyone would achieve the TAS 8-2 FR. In all of these cases a runner of the time redefined what was thought possible to the amazement of the whole community. The result each time is that a new baseline is defined from which others can base their aspirations on. I have no doubt that 4:54 will be achieved – you all have shown a level of passion and dedication to this game that has only grown stronger with time. Congratulations, Niftski!” – darbian

This goes to show how speedrunners regard speedrunning as a community. No one seeks to set a world record for the sake of beating someone on the leaderboards: it’s to push a game to its absolute limits and to show what is truly possible. This message resonates with Kosmic’s comments following his PB: he didn’t care if someone beat his time, as he had accomplished what he wanted. With everyone just cheering for a 4:54, the clock was ticking on history.

But what leap could be needed? At least one framerule was necessary, but the last set of “tricks” were incredibly difficult. This meant that one of them had to be used.

This is where I talk about the 8-1 FPG, which, if done correctly, saves a framerule over the conventional strategy. Due to the length of the stage, an ordinary FPG, which only saves 15 frames, is not enough to save a framerule, as completing the level normally (which has its own share of quirks that is best explained in Bismuth’s first video, which I’ve shared at the end of this blog) gives 0 frames to spare, meaning the level has to be played perfectly. However, it is possible to save a maximum of 19 frames with FPG on this level, although it is significantly harder than a conventional FPG. In order to save the final two frames, players have to perform a frame-perfect “fast acceleration”, a trick that allows Mario to gain ground faster at the start of a level. Miniland would later post a tutorial on how to save this framerule (link here).

With everything now set in motion, we all had to ask: when will history be made?

On April 7, 2021, history WAS made: the final second barrier ever was finally toppled over.

This run was nothing short of spectacular. 1-1 FPG, 1-2 pipe clip, 4-1 FPG, the blazit clip in 4-2, 8-1 FPG, 8-2 TAS framerule, 8-3 FPG, and a good 8-4.

Congratulatory messages came flooding in. AndrewG, darbian, Kosmic, and somewes all offered their congratulations on Twitter, while Taven himself commented “Very speed” on Niftski’s video.

Final Thoughts

As impressive as Niftski’s run is, you’ve probably predicted that the run could still technically be improved by at least a framerule by saving the one in 4-2. While it appears that it may still be a bit out of reach, Niftski has been able to save it in practice using a strategy that was found to be humanly possible by KingOf_JonnyBoy (here‘s his practice run). This would leave the eventual record to be somewhere between 4:54.600 and 4:54.800 (I’m not sure of the timing associated with the Bowser patterns, which is what is used to determine final time). I’ll reiterate the golden rule of speedrunning: the limits are never where we think they are.

Here’s a final breakdown of what is needed for speedrunning Super Mario Bros. (the video above will help put this in perspective):

Getting under 5:00: Perform a wrong warp in 4-2, perform the pipe jump and wrong warp in 8-4
Getting a 4:57: Same as above, but also get BBG
Getting a 4:56: Same as above, but also perform FPG in 1-1 and 4-1
Getting a 4:55: Same as above, but also perform 1-2 pipe clip, Fast 4-2, and 8-3 FPG
Getting a 4:54: Same as above, but also perform 8-1 FPG and “TAS” 8-2

I will also update this blog if information is found to be either wrong or missing in order to best record (pun intended) the history of this illustrious speedrunning category. I leave here a playlist of all Super Mario Bros. speedrun world records that were under 5 minutes, in order. Here’s the record progression, in order (starting with Scott Kessler’s 5:05):

Scott Kessler – 5:05 (set 9/14/2006)
AndrewG – 5:00.721 (set 4/10/2007, first record with glitches)
AndrewG – 5:00.4 (set 12/23/2010)
AndrewG – 4:59.4 (set 12/24/2010, first sub-5 minute record)
AndrewG – 4:58.891 (set 12/15/2011)
AndrewG – 4:58.775 (set 1/15/2013)
AndrewG – 4:58.558 (set 1/19/2013)
AndrewG – 4:58.508 (set 3/21/2013)
AndrewG – 4:58.342 (set 5/19/2013)
AndrewG – 4:58.159 (set 7/1/2013, first record with Fast 4-2)
AndrewG – 4:58.142 (set 10/7/2013)
AndrewG – 4:58.092 (set 3/25/2014)
Saradoc/Blubbler – 4:57.693 (set 6/25/2014, first 4:57, first record with BBG)
darbian – 4:57.627 (set 10/28/2015)
darbian – 4:57.427 (set 1/14/2016, first record with BOTH Fast 4-2 and BBG)
darbian – 4:57.260 (set 4/14/2016)
darbian – 4:57.244 (set 8/10/2016)
Kosmic – 4:57.194 (set 10/3/2016, first record with 1-1 and 4-1 FPG)
darbian – 4:56.878 (set 10/5/2016, first 4:56)
darbian – 4:56.528 (set 10/20/2017, first record with 8-3 FPG)
Kosmic – 4:56.462 (set 2/16/2018, first record with Fast 4-2 via blazit clip)
somewes – 4:56.245 (set 5/25/2018, first record with blazit clip AND 8-3 FPG)
Kosmic – 4:55.913 (set 9/24/2018, first 4:55, first record with 1-2 pipe clip)
somewes – 4:55.796 (set 10/22/2018)
Taven Webb – 4:55.746 (set 8/2/2019)
Kosmic – 4:55.646 (set 1/17/2020)
Niftski – 4:55.430 (set 10/13/2020, first record with TAS 8-2 framerule)
Miniland – 4:55.314 (set 1/2/2021)
Miniland – 4:55.230 (set 2/4/2021)
Niftski – 4:54.948 (Set 4/7/2021, first 4:54, first record with 8-1 FPG, Current WR as of 4/28/2021)

Until then, I hope you all enjoyed reading, and I’d like to extend my heartfelt gratitude and thanks to the speedrunning community, especially the titans that I’ve named in this blog.

Sources: Summoning Salt’s videos