Celeste is not an easy game. Your first time through, you’re gonna die, a lot. As soon as you’ve mastered one mechanic, the game will throw something totally new at you and you’ll be back to dying in the same place over, and over, and over again. It’s not uncommon to rack up thousands of deaths during your first playthrough, so finishing the game at all is worth celebrating. But for some players, that isn’t enough. I’m MKSea, and this is the story of the people who have turned a game about carefully traversing a dangerous mountain, [show early speedrunning] into this [show crazy strats]
The first month (~3:00)
It didn’t take long for Celeste speedrunning to take off. On January 25, 2018, the day the game was released, the first official any% speedrun was already being livestreamed. The runner, who I’m gonna call s-wrecks to avoid being demonetized, beat the game in 1:08:12. s-wrecks had a long history with Celeste already, having been a speedrunner for the prototype version of the game a few years earlier. This version, now known as Celeste classic, was developed in just 4 days by Maddy Thorson and Noel Berry back in July of 2015, and it quickly attracted a small but passionate group of speedrunners. This didn’t go unnoticed by the developers, and as they started work on the full version of the game, they decided to build it with speedrunning in mind. They not only included a built-in speedrun timer, but also designed the game’s mechanics and levels so that most rooms had multiple solutions, giving players the freedom to experiment with faster, riskier Strats if they wanted to. They also recruited 3 Celeste classic speedrunners, s-wrecks, baldjared, and Justin Jaffray, AKA Juice, to be beta testers for the full game. A couple years later, once the game was about to launch, s-wrecks and baldjared created leaderboards and a discord server to build up hype for Celeste speedrunning. The stage was set, and once Celeste was finally released to the public, the race to lower the Any% world record was on.
The game was an instant hit, getting near-universal acclaim from critics and ordinary gamers alike, with speedrunners being no exception. s-wrecks had a bit of a head start as a beta tester, but their day 1 world record wouldn’t stand for long. On January 26, the day after the game’s launch, a runner named OddBod took the top spot. OddBod had been a speedrunner for over a year already, getting top times in Shovel Knight and The End is Nigh, but now, he was going all in on Celeste. He became the first person to beat it in under an hour, with a time of 59:44.
Then, the next day, he set a new record again
and again
But right behind him, another experienced speedrunner was ready to strike.
Hornlitz had been speedrunning for a few years by this point, mostly playing ape escape and various zelda games, holding a longstanding world record in Zelda: Phantom Hourglass. As promised, she snatched the top spot from OddBod, getting a 54:48 on January 27th.
The next day, she lowered the record to 51:44, only for OddBod to strike back with a 49:53 just 6 minutes later. In just 3 days, these 2 runners had lowered the world record by nearly 20 minutes, and they were just getting started. These early runs understandably featured a lot of mistakes, with OddBod’s 49:53 being the first record with under 50 deaths, so there was still plenty of room for improvement. Hornlitz vowed to strike back, but before that could happen, a new runner, Moglar5k, AKA Magolor9000 entered the fray.
Moglar was also an experienced speedrunner, having achieved top times in Undertale, The End is Nigh, and Hollow Knight, and he was ready to add Celeste to that list. His first attempt at an Any% speedrun was just 1 second off of OddBod’s record. Then, a few hours later, he took the top spot with a 49:14, and an hour after that, he got a 47:33. Over the next few days, Moglar improved his time at a crazy pace, pulling ahead of the rest of the pack, and managing to bring the record down below 40 minutes for the first time on February 3.
A few days later, Moglar had lowered the record even further, down to 39 minutes flat, but by this point, competition was starting to heat up. Almost 100 people had submitted Any% speedruns to the leaderboards already, with more joining the community every day, so there were plenty of runners capable of improving the world record even further. On February 9th, fireb0rn, another top-level Hollow Knight speedrunner, briefly took the top spot from Moglar, getting a 38:56, but Moglar quickly took it back just over an hour later.
The next day, CosmicSense, a speedrunner with top times in Cuphead, The End is Nigh, and a bunch of I Wanna Be The Guy fan games, took the world record with a 38:01. Moglar managed to get the record back with a 37:40 a few days later, but this time, he didn’t hold on to it for long. Thanks to a strong lategame and fewer mistakes throughout the run, CosmicSense was able to pull ahead over the next week, bringing the record down to 35:40.
At this point, it had been less than a month since Celeste was first released, and already, 6 different people had held the Any% world record, lowering it by over 30 minutes. And this was just the beginning. But before we go any further, let’s take a closer look at how these runners were able to beat the game so quickly. It’s time for Celeste speedrunning 101.
Celeste Speedrunning 101 (~3:00)
So, you want to try speedrunning Celeste. First things first, you gotta learn the moves. The game only has 3 basic actions, jump, grab, and dash, but by using them in different situations, you can unlock many more speedrunning techniques, or speedtech, for short. Starting with jumping, you have
bunnyhops
WallJumps
Neutral walljumps
and cornerkicks
Next, if you add grabbing into the mix, you get climbjumps
If you dash and then jump, you get
wallbounces
superdashes
and hyperdashes
Superdashes and hyperdashes can both be extended, which involves jumping late in order to gain an extra dash to use midair. They can also be reversed, which involves changing direction while jumping in order to pull off these moves in tight spaces. Finally, if you perform an extended hyperdash starting in the air, then you get what’s called a wavedash.
Once you’ve mastered all of this speedtech, you still have a lot to learn. You need to have a plan for the entire game, memorizing precise strats for every single room you go through. You can’t just wing it in a Celeste speed run and expect to get a top time, after all. But you don’t have to come up with these strats on your own. Even top level speedrunners steal, I mean, uh, borrow strats from other runners, which is a major reason why the world record was improved so quickly early on. Runners of all skill levels worked together to find ways to save time all throughout the game, and this collective effort paid off big time.
Once you’ve picked up solid set of Strats, all that’s left is practice, and then you’re ready to start speedrunning. With the basics out of the way, let’s get back to the world record history.
TGH vs yp ()
2-25-2018
One month after Celeste was released, the top-3 any% speedrunners were neck-and-neck. In first, we have TGH, an experienced speedrunner who found
At the end of the first month, the top 3 Celeste any% speedrunners were neck-and-neck. 1st place was TGH with a 35:32, 2nd place was CosmicSense with a 35:40, and 3rd place was yp with a 35:48. CosmicSense quit speedrunning Celeste after this, so it would be up to TGH and yp to keep improving the world record for the time being. And they wasted no time at all. The very next day, on February 25th, yp set a new record with a 35:27, then, TGH took it back later the same day with a 35:20. The day after that, yp got the record yet again with a 35:15. For the next month, they would continue this pattern, trading world records back and forth again, and again, and again. By the end of March, they had lowered the record all the way down to 32:43, over 2 minutes ahead of where it was at the end of February, and over a minute ahead of 3rd place. So how did they improve so quickly? Did they find a way to skip another large section of a level, or maybe discover some new speedtech? Believe it or not, the answer is neither. They did incorporate new strats, but none of them saved more than a second or two each. Instead, yp and TGH saved time mainly through better execution of existing strats. By playing more aggressively, minimizing airtime, and hesitating less, they were able to save small amounts of time all throughout the game. Each of these improvements was minor, usually a few 10ths of a second each, but put together, they made a huge difference, allowing these runners to pull ahead of the rest of the pack.
At the start of April, yp had to take a break from the world record grind to prepare for Calithon, a speedrunning event where he would do live runs of Celeste and a couple other games. While he was gone, TGH started to pull ahead, bringing the record all the way down to 31:59 on April 8th. When yp returned a week after that, he had a lot of catching up to do.
Without yp to compete with anymore, TGH didn’t need to keep lowering his time to hold on to the record. Other top-level runners were quickly improving, but it would be a while before they could catch up. However, that didn’t matter to TGH. For him, the point of speedrunning was never just to be on top of the leaderboards. Instead, the point was to push this game to it’s absolute limits. The only question was: just how low could the world record go?
The Road to Sub-30 (~6:00 + Montage)
It only took 1 day after the release of Celeste for someone to beat it in under an hour. Then, just 3 days after that, it was beaten in under 50 minutes. A week later, it was beaten in under 40. But 3 months later, it wasn’t clear when, or if, a sub-30 minute run would ever happen. Even if you added up the individual world records for each level in the game, the total was still well over 30 minutes. But if there’s one universal truth of speedrunning, it’s that you shouldn’t bet on anything being impossible, because when you do, someone’s gonna make it their mission to prove you wrong.
[Play montage of TGH WRs, showing clips of his pbs + reactions from 31:38 to 29:59]
After months of grinding, TGH had done it. With hundreds of hours of practice, some new speedtech, and plenty of new strats discovered by the community, he had made the impossible possible. Let’s take a look back at how he pulled it off.
When I went over Madeline’s moveset earlier, the fastest move I covered was the hyper, but as it turns out, that’s not the fastest you can make her go. An even faster option is the ultradash, or ultra. An ultra is performed by first doing a hyper or a wavedash, then quickly dashing diagonally downward. This causes Madeline to maintain high speed for much longer, and when she hits the ground she gains even more speed. If the ultra is performed at the right height above the ground, then it’s possible to gain even more speed by jumping shortly after landing, and this can be extended even further by dashing diagonally downward again. This technique, known as ultra chaining, is only possible in a handful of places in the game, most of which weren’t known at the time TGH got his sub-30. However, he uses plenty of normal ultras to save time throughout the run.
Another variant of the ultra is the grounded ultra. This is performed by first starting a hyper on flat ground, then immediately dashing down diagonally, getting the same speed boost as you would when hitting the ground after a mid-air ultra. Grounded ultras can even be chained together, allowing TGH to maintain very high speed in one room in chapter 5.
Yet another new move TGH used is the corner boost. Corner boosts occur when Madeline is moving toward the top corner of a wall, then quickly grabs the wall and does a climb jump, allowing her to fly past the corner with even more speed than she started with. How does this work? Well, when Madeline hits a wall, the game stores the speed she had during her approach, and if she clears the top of the corner within 4 frames, then the game sets her speed back to the value it stored, as if she never hit the wall at all. But this doesn’t explain where the extra speed comes from. To understand that, we need to talk about jumping. See, jumping gives Madeline a very small speed boost, normally too small to notice. Speedrunners use this in the Prologue by jumping as much as possible, saving over a second across the level, and they also use it after performing wavedashes, hypers, and ultras to give those moves an extra boost, a technique known as bunnyhopping. But what does that have to do with corner boosts? Well, if you are flying toward a wall, and you press grab and jump right before actually hitting it, then the game gives you the same speed boost you’d get if you jumped off of the ground. Combined with the speed stored from Madeline’s approach, this means a properly timed corner boost can give her extra speed.
There aren’t a ton of corner boosts in TGH’s sub-30, but was able to use one to pull off a ridiculous-looking strat in Chapter 5. This strat, known by the community as yeet, involves standing on a block that quickly shoots to the right when Madeline performs a dash. If timed correctly, you can perform a reverse hyper off of this block, giving Madeline an insane speed boost, and if this is followed by bunnyhopping on a nearby platform, then dashing and cornerboosting off the wall at the end of the room, then it’s possible to save multiple seconds in the next room by flying under the moving block. In his sub-30, TGH didn’t quite clear this final block, but he still saved about a second in the previous room from the yeet.
Finally, in Chapter 6, TGH was able to save an additional 4 seconds with a trick known as Lake Skip. Normally, in this room, you need to dash through the water to the right side of the room, then dash into a block, ride it back across the screen, then jump and dash off of it to get to the next room. However, it is possible to skip all of this by grabbing the left wall at a precise pixel, jump off of it, then do a wallbounce off of the bottom corner of the spiked wall to finish the room, saving about 4 seconds. This was discovered back in March by Zkad, but most runners didn’t use it right away because of how precise it was, and failing the trick usually meant dying. However, in May, an update to the game was released that made wallbounces more lenient, making Lake Skip way more consistent, so TGH and other runners started using it in runs.
Even with all of these timesaves, TGH’s 29:59 was still no easy feat. He completed the run with very few major mistakes, so it looked like it would require even faster strats to beat. Over the next couple months, he managed to shave another 18 seconds off of the record, largely in Chapter 5, where he saved time at the end of the level with a move known as a Theo ultra. Theo ultras are ultras or grounded ultras, immediately followed by grabbing Madeline’s trapped friend Theo. When Theo is grabbed, Madeline’s previous momentum is preserved, allowing TGH to quickly jump with the speed gained from the grounded ultra, flying through rooms. But even with this timesave, TGH’s progress was clearly slowing down, and in the mean time, the gap between first and second place was shrinking fast. The community had grown substantially in the last few months, largely due to the publicity it got during Summer Games Done Quick, a popular speedrunning charity event where TGH and yp did an Any% race in front of thousands of live viewers back in late June. By the end of September, 4 runners, Fladervy, Msushi, OddBod, and buhbai had gotten times under 31 minutes, with Fladervy just 26 seconds away from TGH. The competition was heating up, and over the next few months, some of the biggest discoveries in the history of Celeste speedrunning would be made, changing the way the game was played forever.
NOTES
Ultras
Cornerboosts
Dream hypers
The Demodash (~7:00)
On October 14, the first of these discoveries was made by 2 members of the tool-assisted speedrunning community, DemoJameson and Kunogi. They found that it was possible for Madeline to pass through a wall of spikes without dying by performing a precisely timed grounded ultra. This works because certain spike walls are actually treated by the game as separate objects stacked on top of each other, with 4 pixel-wide gaps between them. And as long as Madeline is crouched, as she is during a grounded ultra, then the game treats her as if she is only 4 pixels tall when checking for collisions with spikes and enemies. At first, this trick didn’t seem very useful for speedruns, since it was too precise to be performed consistently, and there weren’t many places where it was even possible. However, later that same day, DemoJameson made another huge discovery. He found that it was possible to pass through a wall of dustbunnies, which work the same as stacked spikes, while dashing in midair. This was done by dashing while holding down for 1 to 4 frames, or less than .068 seconds. You might expect this to just make Madeline dash straight down. Well, as it turns out, the game waits 4 frames after the dash button is pressed to check which direction Madeline should go, so if down is released by that point, and no other direction is pressed, then the game makes Madeline dash in the default direction, which is forward. However, because the down button was held at the start of the dash, the game still treats Madeline as if she’s crouched, making her hitbox just small enough to fit through the invisible gap between the spikes. In chapter 3, this had the potential to save up to 12 seconds, by allowing speedrunners to skip several rooms that were normally required to proceed through the level. Right away, people scrambled to find a way to pull off the trick consistently, and within a few hours, two runners, buhbai and Marlin, independently found a setup. This required players to grab a wall, align Madeline to a specific pixel on the wall, jump, then when she’s at her highest point, dash while holding down for less than 4 frames. It was still a very precise trick, but with enough practice, top-level runners were able to pull it off pretty consistently. People started calling this trick demodash, named after the person who discovered it, DemoJameson, keeping with speedrunning tradition.
Just 2 days later, TGH set a new world record using the demodash, with a time of 29:35. But other runners were catching up fast. Fladervy, who was previously a top-level Shovel Knight speedrunner, had been speedrunning Any% for a few months by this point. By the time the demodash was discovered, he had been on the cusp of a sub-30 for a couple weeks, getting a 30:07, then a 30:05, and even getting 2 30:01 runs on the same day, with the second of these being on world record pace until halfway through chapter 7. With the help of the demodash, he finally pulled it off on October 16, becoming the 2nd person ever to get a sub-30 with a 29:54. 4 days later, he got another personal best with a 29:52, just 17 seconds off the record. This was the closest anyone had come to surpassing TGH in over 5 months, and Fladervy wasn’t about to slow down. After all, the next day, he was scheduled to race against TGH as part of the ongoing Celeste Any% speedrunning tournament. This race was set to be legendary, pitting the 2 best Any% runners against each other, streamed live for the community to see.
As the race kicked off, Fladervy and TGH were neck and neck. Fladervy was a couple seconds ahead after Chapter 1, but TGH took the lead back in chapter 2. In chapter 3, they kept playing almost perfectly, both getting the demodash on the first try. But, in the very next room, TGH made his first big mistake, dying at the end of the room, losing 10 seconds. He played pretty cleanly for the rest of chapters 3 and 4, but Fladervy was on his best pace ever, finishing chapter 4 17 seconds ahead of TGH. He increased his lead even further in chapter 5 and 6, heading into the final chapter with a 24 second lead. Fladervy was playing so well, that he was not just on track to beat TGH, but he was also on world record pace. Was it really possible for someone to break the record during a tournament race like this? Doing so would require a near-perfect chapter 7, which was arguably the hardest level in the speedrun at the time, and the pressure was high. As the commentators and viewers at home held their breath, he entered the level. He started off strong, flying through the first half of the level with near-flawless gameplay. His focus was unbroken, as he pulled off autoscroller skip, key skip, and door skip on the first try. Going into the final checkpoint, it was looking like he might actually pull this off. But the run wasn’t over yet. At flag 9, he missed a tight wallbounce, setting him back 4 seconds. Shortly after that, he missed another wallbounce at flag 4, losing another few seconds. His nerves were surely getting to him, knowing what was on the line, and it was going to come down to the wire. He couldn’t afford any more mistakes. With all this pressure, he made it to flag 3, then flag 2, then flag 1, then...29:31.995. He had clutched out the world record, beating his personal best by over 20 seconds. The community went crazy. Nobody was expecting someone to set a new world record during a tournament race, but Fladervy had pulled off an unbelievable run. For the first time in 7 months, someone other than TGH stood at the top of the leaderboards...for less than a day.
Having finally been surpassed after dominating the category for so long gave TGH the motivation to push himself even further. After congratulating Fladervy on his achievement, TGH tweeted “...You better bet I'll be going all out on Any% the next few days. I've been waiting for this.” Then, less than 10 hours later, he took the world record back on stream with a 29:25. A week later, he lowered it to a 29:21, then 10 days after that, he lowered it yet again to a 29:15. Less than 3 months after the first 29 minute run, a 28 was starting to look doable. And it would happen much sooner than anyone could have anticipated, thanks to another major discovery by DemoJameson.
5B Route (~6:00)
On November 9, 2018, the same day that TGH got his 29:15, DemoJameson discovered a trick that would save even more time in Any% than the Chapter 3 demodash. But to understand it, we need to rewind a few months. On July 29, a speedrunner named revolucion seemed to have discovered a way to unlock a chapter without completing the previous chapter, at least not in the way you’d expect. See, every chapter in Celeste has a hidden cassette tape, and collecting this tape unlocks a special bonus level, known as the B-side of the chapter. B-sides look similar to the normal chapters, or A-sides, as they’re called, except they’re much harder. As it turns out, completing a chapter’s B-side unlocks the next chapter, even if the A-side is unfinished. The Celeste modding community had actually known about this for months, but speedrunners weren’t aware of it until now. So how does this save time? Well, for most chapters, collecting the cassette tape, quitting the level, then beating the B-side is much slower than just beating the A-side normally, so nothing changed. However, chapter 5 was different. The cassette tape could be grabbed in a little over a minute, and its B-side, which I’ll call 5B, could be completed over a minute faster than its A-side, or 5A, so it was theoretically faster to play 5B in Any%. Problem is, 5B is way harder than 5A, not just for casual players, but for speedrunners too. It’s full of way more challenging rooms, and mistakes are way more punishing, since its rooms are much longer on average. Also, the only way to save time using 5B was by skipping several rooms with a precise cornerboost off of a spiked floor. This strat, known as dreadcorner, was so precise and inconsistent that practically no top runners thought it was worth using in Any%. A week later, another, much easier method of saving time in 5B was discovered, using a glitch known as menu storage, but the glitch was fixed a few days later, and although a separate set of leaderboards was created for people who wanted to keep using the glitch by playing old versions of the game, these leaderboards hardly saw any participation, so we won’t include those runs in this video.
This brings us to DemoJameson. On November 9th, less than a month after he had discovered the Demodash, he found yet another way to save time in 5B. This strat, known as bubsdrop, finally made 5B viable for the main Any% category. Bubsdrop is performed in the same room as dreadcorner, saving roughly the same amount of time, but it works much differently. First, you have to proceed through the room and go through the top exit. Then, once Madeline starts moving, you have to quickly tap jump, so that you go back down to the first room. This technique was accidentally discovered a month earlier by buhbai, AKA bubs, in a different room in Chapter 3, which is why the strat was named bubsdrop. After re-entering the first room, if you pause the game and hit retry, you will respawn at the end of the room. And that’s it. Why does this work? Well, it’s a side effect of the way respawning is programmed in the game. Every room is assigned one or more respawn locations, usually one per entrance. When you die, the game checks where you entered the room from, and places you at the appropriate respawn point. However, in the bubsdrop room, there is no respawn point associated with the top entrance, since the developers didn’t expect re-entering the room from there to be possible, so the game instead defaults to the closest respawn point, which is the one at the end of the room. This strat was precise, but it was far more consistent than dreadcorner, so finally, there was a viable way to save time in Any% by playing 5B, shaving off upwards of 20-30 seconds if done perfectly. This was, without a doubt, the biggest timesave discovered for this category in months, and a 28 minute time was right around the corner.
It took TGH less than a week to hit this milestone, first getting a 29:01 on November 13th, then getting a 28:49 the next day. As November wrapped up, TGH lowered the record once again to a 28:45, but at the same time, 2 other runners were catching up. The first was Nero_x2, who previously held multiple coveted super mario 64 world records. At the start of November, his pb was 30:41, and by the end of the month, he managed to get all the way down to a 29:15. The second runner catching up fast to TGH was Msushi, who is a well-known Portal speedrunner and YouTuber. Msushi had been near the top of the Any% leaderboard for a few months already, but after bubsdrop was discovered, he grinded hard, lowering his record from 29:49 down to 28:57, becoming the second person ever to get a 28. But as December rolled around, both Msushi and Nero_x2 switched their focus to other categories, and after lowering the record once more to 28:40, TGH took a break from Any% as well.
What about Fladervy? Well, after briefly taking the world record from TGH in October, he continued running Any% for a bit with no luck, before spending a couple weeks focusing on improving his time in Chapter 5A. He lowered the world record for that level substantially, becoming the first person to beat it in under 4 minutes, but with the discovery of bubsdrop, it was no longer relevant for Any%. This didn’t set him back for long, though. In December, he learned how to speedrun 5B, and he finished the year off with a bang, getting a 28:50 Any% time on the 29th, putting him back in 2nd place, right behind TGH. Then, just over a week later, on January 7th, 2019, he got a time of 28:37, putting him back on top of the leaderboards once again. Over the next week, he lowered it three more times, two of which were on the same day, down to a 28:35. Just as 2019 was getting going, it looked like a new era for Celeste Any% competition was beginning. Over the last 3 months, largely thanks to the discovery of the Demodash and bubsdrop, the world record had been lowered by over a minute, and 5 new runners had gotten times under 30 minutes. This momentum wasn’t about to stop anytime soon, as yet another big timesave was right around the corner.
Prologue Cutscene Skip & The Race to 27 (~4:30 + montage)
For obvious reasons, speedrunners hate cutscenes. No one wants to see the same cutscene over and over again every attempt, especially cutscenes at the start of a game. For the most part, Celeste doesn’t have this problem, since pretty much every cutscene can be quickly skipped from the pause menu, with one glaring exception: The prologue. At the end of the prologue, an unskippable cutscene plays, teaching the player how to dash, taking about 15 seconds. Back in August, when menu storage was rediscovered, speedrunners found out that the Prologue cutscene could be skipped with it, but, as I mentioned earlier, this glitch was fixed by the developers, so speedrunners were still stuck sitting through the cutscene every time they played. However, this changed on January 16, when a speedrunner named cricter discovered that if you paused at exactly the right time, then the cutscene could be skipped. The community was shocked. This was the 3rd big timesave discovered in the last 3 months, and it was right at the start of the game. The devs were notified right away, and they acted fast. Less than 2 hours after the trick was discovered, they announced that they would patch the game to make it even easier, by making the cutscene skippable no matter when you hit pause. Just like that, 15 more seconds were shaved off of the Any% speedrun.
Fladervy jumped on the opportunity quickly, grinding runs the next day, lowering his record 3 separate times, bringing it all the way down to 28:19. Then, a couple days later, he lowered it once again with a 28:16. Just over 2 months after the first 28 minute time, a 27 seemed to be right around the corner. But Fladervy wasn’t the only one going for this milestone. TGH had taken a break from Any% to practice a different Celeste speedrunning category, All Chapters, for the Awesome Games Done Quick charity event, but by the time the Prologue cutscene skip was discovered, he was back on the grind. On January 21, he took the world record back from Fladervy, getting a 28:07. The race for the first 27 was on.
This wouldn’t be an easy feat. Not that getting a world record in this game was ever easy, but as time went on, it only got harder, especially since the big skips discovered recently were both difficult and inconsistent. Both the chapter 3 demodash and bubsdrop required very precise timing, and failing either strat cost more than 10 seconds, essentially killing any chance at a world record. And these were just the most notable examples. Any% was full of plenty of other risky strats that required precise timing, so mistakes were inevitable, and the margin of error for a world record was incredibly low.
Both runners practiced for days on end, and a week after TGH’s 28:07, neither of them had gotten any closer to a 27. A couple days later, on the 29th, Fladervy got a 28:11, including one of the fastest times in Chapter 7 ever performed. This was a fantastic run, but he wasn’t going to be satisfied until he got a 27. Several more days passed without a new world record from either runner, until February 5, when TGH pulled off his best run yet. This run started off strong, with very few mistakes in the early game, finishing chapter 3 almost 3 seconds ahead of his target pace. This continued in chapters 4 and 5, as he hit the halfway point in the run over 7 seconds ahead of pace. Then, a near-perfect chapter 6. Going into the last chapter, he was 14 seconds ahead of the world record. He just needed to stay calm, and avoid making any big mistakes in summit to pull it off. Aside from a death in the 500m checkpoint, he started the level off very clean. His chances at a 27 were looking good, but just one more death, or a few minor mistakes could kill this hope in an instant. He had to stay focused. The 2000m checkpoint was looking clean, except in the last room, where a missed dash left him just pixels away from death. But he caught his mistake in time and kept going. Another checkpoint, another second lost from bunny-hopping slightly too late. But he was almost at the finish line. He just had to stay focused for a few more minutes.
[Play short montage of 3000m clips]
27:55.843. Once again, TGH had made the impossible possible. Just over a year after Celeste was released, the world record had been brought down lower than anyone could have possibly imagined back when it came out.
New Competition (~8:00 + Montage)
When TGH got his 27:55, he and Fladervy were pretty far ahead of the rest of the pack, with 3rd place being a 28:54 by Msushi. But this would soon change. Over the next 6 months, neither of them would improve their times much, and the world record progression came to a screeching halt. TGH had switched his focus to other Celeste categories, as well as other games, only occasionally coming back to Any%, bringing the world record down just a few more seconds to a 27:50. Fladervy, on the other hand, had pushed himself so hard in January that he developed serious hand pain, forcing him to take frequent breaks, making it even more difficult to improve his time. In April, he did manage to pull off a 27:55, but unfortunately, this was in the middle of an unrecorded practice session, so on the leaderboards, his best time was 28:07 from a run he got in February. In the meantime, the rest of the Celeste speedrunning community wasn’t slowing down. After all, the community was still growing, and there were plenty of runners actively working to improve their personal bests.
By late July, 5 runners were within 30 seconds of the world record. In 6th place, there was Marlin, who previously held multiple world records speedrunning The End is Nigh. Marlin had been running Celeste Any% on-and-off for a year, just recently rising to the top of the leaderboards. He got his first 28 minute time in April, then proceeded to lower it a few more times over the next few months, getting a 28:16 in late July. At the time, he also held 3 separate individual level world records, for chapters 1, 3, and 5B. Next, there’s Nero_x2 in 5th, who had an incredible couple weeks in early February where he lowered his pb by over a minute, getting his 28:08 during a race against Fladervy. But after this, he switched his focus to other categories, getting world records in both All Red Berries and All Hearts, never again returning to Any%. In 4th, we have Fladervy, with the 28:07 he got back in February, though as I already mentioned, he had an unrecorded 27:55. In 3rd, there’s Msushi, who grinded Any% several times in 2019, first bringing his personal best down to 28:24 in February, then down to 28:11 in April, then down to 28:02 in July. Finally, in second place, we have Chaikitty, who, like Marlin, had been slowly improving his Any% time since late 2018. In February, he got his first 28 minute time with a 28:42, then proceeded to slowly improve it as the year went on. On July 3, he became the 2nd person to get a recorded 27 minute time, clocking in at 27:57, followed by a 27:53 the next day, just 3 seconds away from the world record. This was arguably the most competitive period in the history of Celeste Any%, as all 5 of these runners, plus TGH himself, had the potential to set a new world record any day.
On July 31, Marlin got a 28:06, climbing up to 4th place. Although this run finished 16 seconds away from the world record, it was actually 7 seconds ahead of record pace after chapter 4 thanks to several new early-game strats. If it wasn’t for 2 failed bubsdrop attempts costing Marlin over 20 seconds, this might have been a world record. But it wouldn’t be long before he had another shot at it.The next day, he got a run that was on a similar pace after chapter 4, and this time, he got bubsdrop on the first try, with no major mistakes for the rest of 5B. With a pretty clean late game, Marlin was able to pull it off, getting a new world record with a time of 27:43. TGH, who had just returned to running Any% again a couple days earlier, got his own 27:43 the very next day, less than half a second behind Marlin’s run. But Marlin wasn’t done. A week later, on August 8th, he did it again, getting an incredible time of 27:34. This run had a solid early game, a near-perfect 5B, and aside from a death late in chapter 6 that cost him 5 seconds, no major mistakes in the late game. After this, Marlin moved on to other categories, but TGH was still going. Over the next week, he kept pushing himself, first getting a 27:37, then a 27:35, then, on August 20th, he took back the world record with a 27:31. He kept trying for a time in the 27:20s for a few weeks, but wasn’t quite able to pull it off, and in late September, he switched his focus away from Celeste to the newly released Zelda: Links Awakening remake. For the next couple months, TGH’s 27:31 stood on top.
By late October, Fladervy had finally gotten a recorded 27 minute time, with a 27:49, and Msushi became the 5th person to reach this milestone, getting a 27:58 and a 27:57. But for the time being, Marlin and TGH stood alone at the top. That is, until Chaikitty came back with a bang on October 27th with a 27:36, 17 seconds ahead of his previous personal best. This run was actually on world record pace going into the final chapter, but TGH’s 27:31 had a near-perfect summit, so just a few mistakes cost Chaikitty the record, at least for now. He switched his focus to the 202 berries category for a couple months, but in December, he came back to Any% once again. On December 11, after streaming attempts for a bit, he was ready to call it for the day, until a viewer named Elvis convinced him to do just one more run. It started out strong, staying within a few seconds of the world record throughout the early game and through the end of chapter 5. In the preceding months, he had practiced his lategame a ton, getting world records in both chapter 6 and 7, so if he played perfectly, he could go as low as 27:19. But it wouldn’t be easy. Chapter 6 started out pretty strong, with a few minor mistakes, but no deaths for the first half of the level. But then, in the short water section, he slipped up, accidentally dashing into a spike, a death that cost him 5 seconds. By the end of the chapter, his best possible time was now 27:27. The summit needed to be nearly perfect for him to set a new personal best, let alone a new world record. He started the level off strong, with a clean 0m section, but two small mistakes cost him a couple seconds at the end of 500m. He was struggling to stay calm, but he still managed to get through the next few checkpoints pretty cleanly. While his live viewers were cheering him on, he entered the last checkpoint: 3000m.
[Show short montage of 3000m].
27:31.346. A new world record by less than 2 10ths of a second. Over the past 6 months, he had gotten within a few seconds of the world record twice, and finally, his hard work had paid off. As 2019 wrapped up, Chaikitty was on top of the Any% leaderboard for the first time.
Compared to 2018, 2019 had been a pretty slow year for Celeste speedrunning. Aside from the Prologue Cutscene Skip introduced in January, no new large timesaves had been used in Any% world records. Instead, runners took advantage of dozens of small optimizations across the game to improve their times. Was the category approaching its limits? Not even close. As 2019 was wrapping up, several top-level runners were changing the way they played Any%, and before long, this change would help bring the world record down to a whole new level.
6B and the race to 26 (~7:00 + montages)
Earlier, I said that, with the exception of chapter 5, it was faster to finish a chapter’s A-side normally than it was to collect the cassette tape, quit the level, and finish its B-side. This was true for a time, until late 2018, when problemstrol and buhbai started working on improving the world record for Chapter 6B. By the end of the year, problemstrol held the record with a 4:25. At the time, the record for 6A was 5:41, and the cassette tape could be grabbed in less than 1:10, so already, it was theoretically faster to play 6B in Any%. So why were no world records set in 2019 using 6B? The main reason is that 6B is a huge step up in difficulty, not just compared to 6A, but even compared to 5B. The level is full of precise falling spike sections, as well as a much harder remix of the boss chase at the end of the level. Also, it is full of long rooms, so there’s a high risk of losing 10 or more seconds by dying near the end of a room. So, since playing 6B would only save 5-10 seconds for top runners if played perfectly, very few people bothered to do so in Any%. One notable exception was revolucion, who, a few months earlier, had submitted an Any% run with a time of 31:27 using both 5B and 6B, before bubsdrop was even discovered.
6B continued to see very little adoption through most of 2019, but as the year was wrapping up, several top-level runners, namely Psyched, buhbai, Auroradash, and Evansfight, started to use it in runs. However, the first person to set a new world record using 6B would be Marlin, who marked his return to Any% with a 27:30 on January 10th. In this run, he actually lost time in chapter 6, since he died 3 times in 6B, and he lost 10 more seconds in the summit. This was only a world record because of his blazing fast early game, where he used new, faster strats to finish chapter 4 18 seconds ahead of the previous world record. Just 3 days later, Chaikitty took the record back with a 27:29, then lowered it a week after that to 27:28.949, without playing 6B in either run. Chaikitty wasn’t able to keep up with Marlin’s pace in the first half of the game, but he made up for it with very fast times in Chapters 6 and 7. A few days later, on January 23, Marlin finally got a run that saved time with 6B. As usual, he started out with a strong first half, finishing chapter 5 with a total time of 13:32, a few seconds ahead of his previous record. He lost 4 seconds to a failed Lake Skip at the start of chapter 6, losing a few seconds, but he made up for it with a much cleaner 6B. Going into the last chapter, his best possible time was 27:03, so he had the potential to smash the record. But it wasn’t meant to be. The first half of summit was very clean, with just a few small mistakes, but 3 deaths in the last 2 checkpoints cost him over 15 seconds, and he finished the run with a time of 27:22. This was still a huge world record, but it was clear that Marlin had the potential to go much lower.
Chaikitty ended up improving his personal best 2 more times, getting a 27:26, followed by a 27:23 a couple weeks later, with the latter run featuring a new summit world record. But this was the lowest he would go for now, as he switched away from Any% for the time being. In the mean time, Marlin kept setting new world records. On Feburary 1st, he got a 27:18, then, on the 10th, he got a 27:14. But at the same time, after several months away from Any%, TGH returned.
[Show short montage of TGH pbs from twitter screenshots/clips up to 27:12]
Within a couple weeks of returning to the category, TGH once again had the world record, but less than a day later, Marlin took it back with a 27:10. These 2 runners were neck and neck, and they were both working toward the same goal: becoming the first person to beat Celeste in under 27 minutes. It had been over a year since the first 27 minute time, and in that time, top-level Any% speedrunning had gotten a whole lot harder. The strats had gotten more complicated, more precise, and more risky. Even if you could manage to get through the first half of the game on a good pace, then there was a good chance that 6B would kill your run. But TGH and Marlin had been doing this for a long time, and they weren’t about to back down from a challenge. The race to 26 was on.
Marlin had gotten less than a second away. If it wasn’t for the death at flag 9, this would have been the run. But just 8 days later, he topped it.
[Show flag 1 of 27:00.117 run]
A tenth of a second away. This was getting ridiculous. But Marlin wasn’t going to let this get him down, and just 4 hours later, he got this run.
[Show flag 1 of 27:00.015 run]
.015 seconds, less than a frame away from getting the first 26. Over the course of 8 days, he had gotten 3 separate runs less than a second away from the milestone he had been working toward for months. You can’t make this stuff up. But even after all this, he didn’t give up. On April 1, he got yet another run that looked like it might be the one. Coming out of Chapter 6, he was 7 seconds ahead of the world record. He just had to lose less than 15 seconds in the summit. 0m, no deaths. 500m, no deaths. 1000, 1500, 2000, and 2500m, no deaths. He had lost a few seconds from minor mistakes, but going into the final checkpoint, he still had 8 seconds left to lose.
[Show montage of 3000m]
26:57. After getting painfully close 3 separate times, Marlin had become the first person to beat Celeste in under 27 minutes, the first minute barrier to be broken in over a year. Finally, he could end his Any% grind, proudly standing on top of the Celeste Any% leaderboards.
TGH kept trying for a 26 for a while longer, and he got close a few more times, but he was beginning to burn out. After holding the world record so many times, the pressure to stay on top was enormous. People would come into his streams, expecting to see perfect Celeste gameplay, only to be disappointed by the ups and downs that all speedrunners experience. The pressure got to be too much, and he stopped trying to take back the Any% world record. He continued to speedrun other categories in Celeste for a while longer, but eventually, he moved on to other games, leaving this chapter of his life behind him. After dominating Celeste Any% speedrunning for the better part of the last 2 years, setting over 50 world records in the category, his final time would stand at 27 minutes and 2 seconds, marking the end of an era.
Buhbai
TGH wasn’t the only top runner to abandon Any%. After Marlin got his 26:57, top-level competition in the category came to a screeching halt. For the next 7 months, none of the runners in the top 5 would improve their Any% time. They all either moved on to other categories or quit Celeste speedrunning entirely. Did this mean the community was dying? Not at all. In many ways, the community was more alive than ever. New people were trying out Celeste speedrunning for the first time, and plenty of runners were getting their first 29, 28, and 27 minute times. But no one was going for a 26. After 2 years of competition, the strats required to compete for a world record were becoming so precise, so difficult, that hardly anyone was up to the task. That is, until the 4th place runner at the time, buhbai, returned to Any%
buhbai had been one of the best overall Celeste speedrunners since the game came out, never straying far from the top of the leaderboards. He held many world records for individual levels, and he was known for using strats that even other top runners thought were too risky. However, any% had never been his main focus. Instead, he spent far more time playing the All Red Berries category, which he absolutely dominated. Despite this focus, he still managed to stay near the top of the Any% leaderboard for years. Once or twice a year, often just as a break from All Red Berries, he would take a couple weeks to play Any%, get close to the world record, then go back to whatever he was doing before. While these detours never lead to a new world record, he got close several times. Back at the start of 2020, over the course of about 2 weeks, he managed to improve his time from 28:39 to 27:32, getting less than a second away from the world record at the time. But it would take more time and more focus to get over the finish line. In November 2020, 7 months after Marlin got his 26:57, he decided to give Any% his full focus for the first time ever.
He got his first pb on November 17, pulling off a time of 27:13, nearly a 20 seconds improvement over his previous best time. But this was just the beginning. If we take a look at how this run compared to Marlin’s world record, we can see that buhbai was about 5 seconds ahead out of Chapter 3, and almost 10 seconds ahead out of Chapter 4. How did he save all this time? Well, a lot of the timesave in chapter 3 was a result of making fewer mistakes, but some of it came from new strats. None of these strats were game changers on their own, each saving less than a second, but put together, they saved at least a couple seconds, and at this level, every second counts. Chapter 4, on the other hand, was a much different story. The majority of the timesave in this chapter came from a single strat, a strat that had been known about for at least a year, but was so precise and inconsistent that basically no top runners were willing to try it until now. This strat, which saves 6 seconds, is known as 2nd blockless.
2nd blockless is one of a group of strats known as autoscroller skips. See, Chapter 4, as well as the corresponding section in Chapter 7, has several rooms that normally require riding a slow moving block through several obstacles. These rooms, known as autoscrollers, are a big pet peeve for speedrunners, as they typically have no easy method of completion that bypasses the moving block. There are 4 autoscrollers that Any% speedrunners encounter: 2 in Chapter 4 and 2 in Chapter 7. By the time buhbai started his Any% grind, the community had discovered methods to complete one of the chapter 7 autoscrollers and both of the chapter 4 autoscrollers without riding the moving blocks. The first of these can be skipped by performing a hyper bunnyhop, then wall kicking over the floating platform, grabbing a small corner, then performing a trick known as a ceiling pop to essentially jump under the bottom corner, then dashing and corner kicking off of the bottom right corner. Although this is a very precise trick, it is made fairly consistent thanks to a setup discovered by buhbai himself back in late 2019. Top-level any% speedrunners