The answer
Skyborne
| Place | Race | Scenarios won | Range | Racial worth |
|---|---|---|---|---|
| #1 | Skyborne | 73% | 45–82% | +2.4% |
| #2 |  Human | 16% | 9–36% | +1.7% |
| #3 |  Gnome | 11% | 9–19% | +1.6% |
Scenarios won: in how many of the fights each race did the most damage. Range: where that share lands if the one unconfirmed game value goes either way (the deciding question, below). Racial worth: damage gained over a mage with no racial.
Every set of shares on this page is rounded to whole percents that add up to 100.
- Why Skyborne: Wind Blessed, 3% faster casting, always on.
- If casting speed doesn't shorten the global cooldown in WoW Forever, Skyborne still wins 45%. We test it on launch day.
- How much it matters: about 2.4% more damage. Pick what you love.
What a scenario is
A scenario is one raid boss fight, simulated. We built 10,000 of them, and each one draws at random:
- how long the fight lasts: 3 minutes in the middle, from 45 seconds to 10 minutes;
- the raid: 10, 20 or 40 players, and the buffs that come with it (Kings, Mark of the Wild, Prayer of Spirit, Moonkin Aura, Curse of the Elements, Mana Spring, Judgement of Wisdom);
- consumables;
- your gear, from fresh level 60 to raid-geared (300 to 750 spell power);
- how much of the fight you get to cast, and how much the boss resists;
- every game value nobody has confirmed yet (listed in the full method).
All three races play every scenario, with the same draws. The race that does the most damage wins it. Within a scenario the races differ only by their racials.
The fight: one boss, one target, level 60, talents 32/3/16 (Arcane / Fire / Frost). The rotation: Arcane Blast to build stacks, then Arcane Wrath (with Presence of Mind), Arcane Missiles on Missile Barrage, and Frostbolt while the stacks fall off. Mana is counted in full.
Built from the game, not from guides. Spells, talents and racials are read from the game's own tables for build 1.60.1.70450. Random seed 20261104, so every number here can be rerun exactly. Sampling error: ±0.9 points.
Why Skyborne wins
Each race's racials as build 1.60.1.70450 has them, and what they're worth to a Arcane Mage: the damage gained over a mage with no racial, in the middle scenario, and the spread across 8 in 10 scenarios.
| Race | Racials | Worth |
|---|---|---|
| Skyborne | Wind Blessed: 3% faster casting, always on. Read Leyline: a 2-second cast for Energized, double mana regeneration for 30 seconds, every 2 minutes (15 minutes near a ley line). Worth casting in 25% of the scenarios. Elemental Insight: +5% damage against Elementals. | +2.4%1.3–6.2 |
| Human | Ruthless: +2% crit chance, always on, with any weapon. The Human Spirit: +5% Spirit. | +1.7%1.6–2.3 |
| Gnome | Eureka!: for 8 seconds every 2 minutes, spells cost 10% less and hit 10% harder. Expansive Mind: +5% mana. | +1.6%1.2–2.6 |
Faster casting is more spells in every fight, and Wind Blessed is always on.
The shortest fights go another way: at 30 s, Gnome wins 54%. From about 45 s on, Skyborne leads.
The numbers as a table
| Fight length | Skyborne | Gnome | Human |
|---|---|---|---|
| 30 s | 46% | 54% | 0% |
| 45 s | 68% | 29% | 3% |
| 1 min | 65% | 19% | 16% |
| 1 min 30 s | 73% | 11% | 16% |
| 2 min | 74% | 11% | 15% |
| 3 min | 70% | 12% | 18% |
| 5 min | 77% | 2% | 21% |
| 7 min | 79% | 1% | 20% |
| 10 min | 82% | 1% | 17% |
The deciding question: does casting speed shorten the global cooldown?
After every spell a mage waits out the global cooldown, 1.5 seconds. Instants, Scorch and Arcane Blast are all held back by it. If casting speed shortens it too, Wind Blessed's 3% makes every one of them faster; if not, it only speeds up the longer casts. The game files don't say. Both open simulators speed it up, after a report from the mage community, so the model says yes in 75% of the scenarios.
If it does
- 82% Skyborne
- 9% Human
- 9% Gnome
If it doesn't
- 45% Skyborne
- 36% Human
- 19% Gnome
Skyborne stays first either way: 82% if it does, 45% if it doesn't.
How we test it
On launch day, 4 November: a Skyborne mage casts instants back to back on a target dummy, and we read the time between them in the combat log. Under 1.5 seconds settles it. This page gets the answer, dated, in its changes.
How much it matters
Skyborne's racials are worth about 2.4% more damage than none. The others' are worth 1.7% (Human) and 1.6% (Gnome).
In a single pull, it's close. With the same raid and the same gear, a Skyborne out-damages a Human in 54% of pulls (45–69% across the scenarios). We count the luck of hits, crits and partial resists; real pulls have more, so the real number is even closer to 50.
Pick what you love. The gap is real, but small.
The human factor
Simulators press every button the moment it's ready. People don't. Human's racial is always on; Skyborne's Read Leyline and Gnome's Eureka! have to be pressed. With perfect use Skyborne wins 72.8% of the scenarios. With less:
| How often Read Leyline and Eureka! get pressed | Skyborne | Human | Gnome |
|---|---|---|---|
| Every time | 72.8% | 16.0% | 11.1% |
| 3 times in 4 | 72.4% | 21.9% | 5.7% |
| Half the time | 70.7% | 26.4% | 2.9% |
| 1 time in 4 | 69.3% | 29.3% | 1.3% |
| Never | 68.4% | 30.8% | 0.8% |
In the middle scenario, Skyborne has to press Read Leyline 22% of the time just to draw level with Human. After launch we count how often our own raiders press theirs.
Why other guides disagree
The open simulator's Arcane table, gunba/wow-forever-sim (a fork of wowsims/classic), ranks Gnome, Skyborne, Human. So we gave our engine the simulator's own character sheets and its fight (5 minutes, no movement) and ran it two ways: once with Skyborne's 1% faster casting, which is what its numbers imply, and once with the game's 3%.
| Against Human | Simulator | Ours, Skyborne at 1% | Ours, at the game's 3% |
|---|---|---|---|
| Human | 724.8 DPS | base | base |
| Gnome | +1.01% | +2.96% | +2.96% |
| Skyborne | +0.10% | +1.42% | +2.03% |
| Order | Gnome, Skyborne, Human | Gnome, Skyborne, Human | Gnome, Skyborne, Human |
Same order with its inputs: the two engines agree. With the game's 3%, Gnome still leads on its sheets. Those give each race its own gear (the simulator searches gear for each race separately); in our scenarios every race wears the same gear, so only the racials differ.
We read the simulator's published table and didn't run it. It's open source under the MIT license.
The full method
Every unconfirmed game value, and the odds we gave it
Each scenario draws every one of these. The ranges on this page are how far a share moves when the deciding question goes one way or the other; it's the one that moves the answer most.
| Value | In the scenarios | Why it's unknown |
|---|---|---|
| Casting speed shortens the global cooldown | yes in 75%, no in 25% | the game files don't say; both open simulators speed it up for every spell, after a report from the mage community; nobody has timed it in WoW Forever yet |
| Casting speed speeds up channels (Arcane Missiles, Evocation) | yes in 10%, no in 90% | a logged mage's missiles stayed one second apart with a casting-speed buff of the same kind |
| A ley line is near the raid boss | yes in 15%, no in 85% | with one, Read Leyline's Energized (double mana regeneration) lasts 15 minutes; without, 30 seconds every 2 minutes; where ley lines are isn't in the game files, and raids are the least likely place |
| Energized adds to Evocation's regeneration | yes in 50%, no in 50% | both are the same kind of effect, which normally adds up, but one may replace the other |
| Ice Lance's spell power coefficient | 0.05 to 0.2 | the game files say 0; public beta numbers fit about 0.10 |
| Improved Scorch's Fire Vulnerability counts as one of your Fire effects | yes in 70%, no in 30% | it's a Fire effect of the mage's own, like Ignite (the open simulator counts it) |
| MP5 on gear pays every second | yes in 15%, no in 85% | the open simulator's unconfirmed option |
| The raid boss is an Elemental | yes in 5%, no in 95% | Skyborne's Elemental Insight adds 5% damage against them; the bosses' creature types aren't in the game files we read |
How the model reads the game
The model reads the game's own tables (DB2) for build 1.60.1.70450, exported from wago.tools: every level-60 mage spell at its highest rank, WoW Forever's talent tree, each talent's value from its own curve, and the racials. The spells' data decides what modifies what: which spells Eureka! makes cheaper and harder comes from the spell's own flags, not from its tooltip. Nothing is a number copied from a guide.
Procs (Heating Up, Fingers of Frost, Missile Barrage) and Combustion's crits are carried as chances, not rolled, so a scenario's damage is its expected value. The head-to-head pulls roll the dice.
10,000 scenarios with the random seed 20261104, so every number here can be rerun exactly. The fight-length chart reruns 600 scenarios at each of 9 fixed lengths.
What we typed in by hand
- The talent build (51 points): Arcane 32 / Frost 16 / Fire 3, the best of the builds we compared.
- The rotation, a raider's usual priority: Fire keeps five Fire Vulnerability stacks with Scorch, casts Pyroblast on Heating Up and Fireball otherwise; Frost casts Ice Lance on Fingers of Frost and Frostbolt; Arcane builds Arcane Blast stacks, then Arcane Wrath, Arcane Missiles on Missile Barrage and Frostbolt. Every spec uses its mana gem, potions and Evocation when the mana runs low.
- The races' stats against the mage baseline: the game keeps them in tables wago.tools doesn't export, so the values are wowsims'.
- Spell hit against a raid boss (83% base, 99% at most) and crit from Intellect: Classic's rules.
What isn't in the model
- Set bonuses, trinkets and procs from gear: they help every race alike.
- Fights with more than one target, and movement beyond how much of the fight you get to cast.
- Power Infusion from a priest.
- PvP. The PvP racials don't count here.
The data. The rankings, ranges, racials' worth and the fight-length sweep on this page as files, free to reuse with credit (CC BY 4.0): CSV and JSON.
Our prediction, before launch
Simulated on 19 October 2026, on build 1.60.1.70450, before WoW Forever launches on 4 November. After 9 December we grade it against our own raid logs, here on this page.
Changes
- Simulated and written, on build 1.60.1.70450.