Talents that define heroes
The new talent system has reinvigorated the game. While for the majority of heroes it simply allowed to keep up with the game’s progression and smooth out the power level kinks, for some, it has completely changed the way the heroes could be potentially played.
Level 25 talents come into play more often, than initially assumed. The changes to the XP and levelling mechanics, coupled with the meta changes, ensured that at least the core heroes get to them in the majority of games.
While for many the level 25 talents still remain uninspired, yet effective, some heroes got unique bonuses that can’t be bought with gold. The most effective of them are not necessarily the most interesting, but they certainly change the hero enough to make a massive difference, especially when compared to the alternative.
Templar Assassin
While we would generally refrain from comparing the Respawn Time Reduction (RTR) to the alternative in the future cases, this one is special: most RTR alternatives win by 4-6%, yet in case of Templar Assassin the difference rises up to 9.5%, signifying that not only is the RTR a generally poor choice for the vast majority of heroes in most situations, but the +3 Refraction Instances is a very good talent on its own.
Starting in the 4k bracket, the hero already wins more games, than she loses. Her situation further improves in the 5k+ one, where she reaches her peak of almost 56% winrate. Moreover, she is one of the heroes who is consistently above the curve when it comes to winrate-level relationship:
In general, when you pick this hero matters a great deal—she excels in games where no enemy can quickly burst through her shields, allowing her to remain safe, fully utilize her 
Blink Dagger and remain on the offensive once her team takes control of the tempo.
As such, it is understandable why her winrate drastically increases in the higher brackets—there is no denying that she requires a certain amount of skill to be played well, but on top of it, better players tend to understand when to pick the hero in the first place. And in games where dealing with Refraction is already complicated, having three extra layers can make the hero almost unkillable.
Skywrath Mage
+14 mana regen or -4s Ancient Seal Cooldown? The answer to this question surprisingly favors the extra regeneration. By a significant margin of 6.4% winrate.
Mana regeneration talents tend to be heavily undervalued, while in most cases being a better choice for your average game of Dota. Even smaller increases tend to translate into massive winrate differences. And in many cases the “why” becomes obvious after some thought—using spells is good and using them more frequently is even better.
+14 mana regen means that Skywrath Mage essentially gets a free Arcane Bolt every 5 seconds. Over the course of the fight, it will generally result in 300+ extra mana and will ensure a full mana bar for the next engagement. Moreover, it will also allow the hero to spam his abilities to simply farm and push—at this stage of the game one Arcane Bolt can frequently be enough to take down a creep, since the ability scales really well into the lategame.
Skywrath Mage is also one of the few supports, who actually has higher chances of winning, as the game progresses:
Granted, the sample size for the level 25 supports is generally rather small, but the trends are there. 60+ minutes into the game most support heroes have a much smaller impact on the outcome of the game, but not Skywrath Mage, super-charged with “infinite” mana.
Lich
A free 
Eye of Skadi is a massive argument in all engagements. The -30% MS and -30 AS slow from attacks fully ignores spell immunity, fully stacks with other unique attack modifiers and can completely ruin the day of the enemy melee cores.
It does have a number of restrictions, obviously. Lich has a smaller attack range of 550, forcing him to be in the center of the action if he wants to contribute his auto-attacks, while low agility gain and high attack animation duration also prevent the hero from becoming a complete kiting nuisance.
On the bright side—out of most supports, Lich has the highest chances of reaching level 25. He has a decent ability kit, which ensures his utility at all stages and can stall the game. On top of it, he can Sacrifice his own creeps, boosting his XP gain.
He is also very item-independent, allowing him to concentrate on wards and other support items in the early game. In the later stages, however, given his talents, he can transition into a decent damage dealer with a sudden half-a-Rapier talent at level 20 and a sudden Eye of Skadi talent at level 25.
Morphling
+400 Waverform Range seems like a rather mediocre talent, especially when compared to a decent alternative of +50% Replicate damage, yet the former wins by an almost 6% margin.
With talents like this, what people tend to not realise, is that it has far more consequences, than simply added mobility. With a travel speed of 1250, extra 400 range corresponds to extra ~0.3s of invulnerability, which can potentially translate into an extra “free” auto-attack or an extra cast. While seemingly minor, these small “gimmicks” is what makes Dota this incredibly complex and beautiful game.
Given the damage the hero can output, having a small edge in terms of both DPS and Survivability can lead to massive consequences. It doesn’t mean that the mobility aspect should be disregarded—on the contrary, it is the small pieces all coming together that trample a potential enemy core illusion with 130% base damage output.
Puck
+75% Illusory Orb Speed/Distance talent is annoying to deal with. It super-charges the mobility aspect of the spell, resulting in a massive 3400 range escape/initiation tool. On top of it, it makes landing the nuke aspect of the ability a triviality, even from a very high range.
Catching Puck in the later stages of the game can become simply impossible—there are very few heroes who can catch up to this trickster. That said, the hero really needed that upgrade. His utility and damage in the late-game is not impressive by any means, and he fully depends on his items to stay relevant. As such, dying is not an option for this hero—he needs to keep on farming, getting items and being a split-pushing/wave-clearing nuisance to have a full impact on the game.
Closing Thoughts
Given the amount of changes the talents got two months after their initial release, it is probably still being tweaked and developed internally by Valve. There is a massive amount of variables to consider and making the system truly work will definitely take time. For now, in some cases, talents are actually inferior or comparable to the attribute boost option of the past, which is neither effective, nor interesting.
Yet, given the talents discussed today, the system has potential. Having upgrades, which do not necessarily rely on the previous game mechanics, but instead provide some unique bonuses is possible. Let’s hope the development of the game does not stop here and more heroes will get equally appealing alternatives that will define them.
isnt puck a girl?
Last, probably
ooooooh boy is the 50th mine?
52 bit*h
I feel like Windranger would be a lot better at the spellcasting role if she got a much higher mana regen like skywrath has. Power shot is just not great, but if you can spam it, it's far more useful.
54th. I agree about windrunner's mana requirements
@time I don't think you really need to sunder more than once per fight. Plus, if you get the stats, more HP, more dmg and attack speed=less need(or targets) to sunder. If you're in the middle of the fight as TB(where the 10 second sunder might sound better), then you're playing a very risky game. Unless you're playing a 4 protect 1 game, and just keep sundering the life from your teammates as you stay in the you fight in the middle of everything.
@KawaiiSocks pls find a way to remove the people who don't contribute to the discussion, and just keep saying "1st!" and such.
I hate to say, but i dont like this kawaisock's opinion, it's not very accurate
Nice
1st, 2nd, 3rd.... idiot'st, Why do moderators allow nuisance coments?!
Puck is a female.
Sixtieth