Published July 19, 2026
CosmicDrift.io shows you a level number in the HUD — "LEVEL 347 · APEX TIER" — and it's easy to assume that's a persistent account stat, the way an MMO character level is. It isn't. Level resets to 1 every time you die. What's actually going on underneath is more specific than "score," and understanding it changes how you think about a run: it's a single accumulating mass value, tracked only for your current life, run through a lookup table that was deliberately built to feel different at different points in the climb. This post is about how that table actually works, why it's shaped the way it is, and where the design is still first-pass and admittedly untested at real scale.
There's no "level" field that goes up when you hit a milestone. There's a single number — call it your mass score, tracked internally as orbs collected — that goes up every time you eat an orb or land a kill, and your level is whatever that score maps to on a lookup table. Die, and the mass score resets to zero for your next life. Level is a read-out of that score, not an achievement that sticks around once you've reached it. This is the same underlying number that drives your stardust earnings and your visual size in that life — one accumulating value, three different things reading off it.
The universal kill bonus matters here too: every kill — bot or player, regardless of which perks either side has — grants a flat +10 mass on top of whatever orbs you'd already collected, the same units orbs use. A string of kills adds up fast against the table below, faster than it might look if you're only tracking orb count in your head.
The old version of this system (before the 1,000-level rework) used one formula for the entire curve — mass required scaled with the square of the stage number, the same rate of increase from level 1 all the way to the top. That's a real design problem, not a cosmetic one: a single power curve forces one number, the exponent, to do two jobs that pull in opposite directions. Make it low enough that early levels feel fast and rewarding in your first few minutes, and the late game becomes trivially cheap — anyone who plays for an hour blows through the "impressive" levels with no real grind left. Make it steep enough that the late game stays a genuine grind worth bragging about, and the first ten levels — the ones that convince a brand-new player the game has momentum — turn into a slog before they've had a chance to decide they like it.
There's no exponent that solves both problems at once, because they're not the same problem. So the v3 system throws out the single formula and replaces it with a 101-point anchor table — fixed mass requirements at stage 0, 10, 20, all the way to stage 1000, with the levels in between filled in by straight-line interpolation between their two nearest anchors. That's the actual trick: because it's a table, not a formula, the rate of mass required per level can change from decade to decade. Early decades can demand very little; middle decades can demand a lot more; and the table can flatten out entirely near the end. A single curve can't do that. A table can, because every ten-level segment is free to have its own slope.
"No single power curve can make level 100 feel exciting in a 15–30 minute session without also making level 1000 trivially cheap — the anchor table exists specifically to let the per-level rate change decade by decade instead of being locked to one exponent for the whole climb."
Reading across the anchor table, the mass curve breaks into three distinct phases, each with its own intent:
This phase is tuned for the player who's never touched the game before. The mass required per level starts small — roughly 28 mass per level through around level 10 — and climbs steadily as you go, reaching about 450 mass per level by the 100–110 range. The target is a player reaching level 10 within about a minute of playing, and level 100 within their very first session. Most players never see past this phase, so it's front-loaded on purpose.
Past level 100, the per-level rate keeps climbing every ten-level decade — roughly +50 mass-per-level added each decade, from 450/level at 100–110 up to 2,450/level by 500–510. This is the phase for players who kept playing past the first session: it's a genuine, escalating cost, not a flat continuation of the early pace.
Rather than let the rate keep climbing for another 490 levels into an ever-steepening wall, the table freezes the per-level rate at 2,450 mass/level — the same rate as the 500–510 decade — for the rest of the climb to 1000. The reasoning: fewer than 1% of players will ever reach level 500, and an ever-steepening curve would make level 1000 effectively decorative — mathematically present but functionally unreachable for anyone, no matter how dedicated. Freezing the rate keeps the long tail a flat, predictable grind instead of an impossible one, so a sufficiently dedicated player can actually get there.
Here's a representative slice of the actual anchor data, straight from the game's constants, alongside the visual size multiplier (more on that below) and a rough sense of how each level feels to reach:
| Level | Mass required | Bot scale | Rough feel |
|---|---|---|---|
| 2 | 50 | 1.14× | a few seconds |
| 10 | 250 | 1.32× | about a minute |
| 25 | 1,500 | 1.50× | a few minutes |
| 50 | 5,500 | 1.71× | under 10 minutes |
| 100 | 21,500 | 2.00× | reachable within a typical session |
| 250 | 141,500 | 3.13× | a long, dedicated single session |
| 500 | 591,500 | 5.00× | multi-hour grind (scale cap reached) |
| 750 | 1,204,000 | 5.00× | flat — no further visual growth past level 500 |
| 1000 | 1,816,500 | 5.00× | flat — server-wide rare event, same size as level 500 |
One thing worth being upfront about: this table doesn't account for death-scatter mass. A bot killed by a player can drop up to 1,200 mass worth of orbs in one spot; a bot that dies ambiently to another bot drops up to 60. A single well-timed kill can be worth over 11 minutes of ambient orb-collecting in one pickup, so real progression in a populated multiplayer room — especially for someone actively hunting instead of just grinding orbs — runs meaningfully faster than the raw table implies.
It's worth being precise here: the "Bot scale" column above isn't derived from the mass table at all. Visual size — how big your entity actually is on screen — comes from a completely separate formula keyed directly on level number, not mass. It's a three-tier piecewise curve:
Levels 1–100 use a front-loaded power curve running from 1× up to 2×. Levels 100–500 switch to a straight linear climb from 2× up to a hard cap of 5×. Levels 500–1000 are completely flat at that 5× cap — no further visual growth at all, no matter how much further you climb.
That cap is a deliberate, direct redesign of an earlier version. The original size formula was a single power curve that topped out at 22× scale — and a player who actually got pushed to level 250 in testing measured out at a 9.69× scale, which was reported directly as making the game literally unplayable for that entity: too big to see around, too big to even attempt fighting because the head itself dominated the screen. That same old curve also front-loaded so much of its growth into roughly the first quarter of the level range that the remaining three-quarters were nearly visually flat — the same "one number can't do two jobs" problem the mass table above exists to solve, showing up again in a different system.
The redesign was built around a simple, honest piece of reasoning about the actual player population: the large majority of players die or stop playing within the first 50–100 levels and need to see visible progress fast, while fewer than 1% will ever reach level 500 or beyond. So growth is front-loaded for the levels almost everyone actually plays, a controlled linear climb carries the "dedicated player" range, and then it goes flat past level 500 — at which point raw size stops being the reward. Benefits for reaching level 500+ that aren't just "bigger" are on the roadmap but not built yet.
In the interest of not overselling this: both the mass-anchor table and the scale-curve constants are documented internally as first-pass numbers, meant to be retuned after real playtesting shows how fast players actually accumulate mass and how the size spread actually feels in a populated room. The design goals above — three phases with genuinely different feel, a size cap that keeps high-level play actually playable — are the parts that were reasoned through deliberately. The exact numbers that implement those goals (28 mass/level early on, +50/level-rate per decade, a 5× cap specifically instead of 4× or 6×) are best-guess starting points, not numbers pulled from data. If a decade turns out to feel too slow or too fast once real players are grinding through it, the anchor table is exactly the kind of structure that's easy to retune locally — you can adjust one decade's slope without having to re-derive a global exponent that affects every other level too.
Level isn't a persistent stat — it's a read-out of a mass score that resets every life, mapped through a 101-point table instead of one formula specifically so the early game can feel fast while the late game stays a real grind. Every kill adds a flat +10 mass on top of orbs. Visual size is a separate, deliberately capped curve that front-loads growth for the players who'll actually see it and stops entirely at 5× past level 500, replacing an earlier version that made high-level entities unplayably huge. None of the specific numbers are claimed to be final — they're a reasoned starting point, waiting on real playtesting to prove out.