Four AI estimates compared
Claude, Gemini, Grok, and ChatGPT were each asked how a Last Z account should be balanced. Here is what each said, side by side. None of it is official or verified.
How power should be split
Claude, Gemini, and Grok each gave target shares of total power.
| Category | Claude | Gemini | Grok |
|---|---|---|---|
| Troop | 40% | 35–40% | 25–32% |
| Hero | 28% | 25–30% | 18–25% |
| Structure | 15% | 10–12% | 28–35% |
| Tech | 12% | 15–20% | 12–18% |
| Mod vehicle | 5% | 5–8% | 5–10% |
Claude gave one number per category. Gemini and Grok gave ranges. Grok’s ranges have midpoints that add up to 104%, so they cannot all be hit at once.
ChatGPT’s different approach
ChatGPT did not give power shares. It gave weights for scoring how developed an account is. Its files say these are not a spending plan or a power split, and that displayed power can overstate how strong an account is in battle. It also lists “Buildings” where the others list “Structure”, which is only a rough match.
| Category | Scoring weight |
|---|---|
| Heroes and equipment | 30% |
| Technology | 25% |
| Troops | 20% |
| Vehicles | 15% |
| Buildings | 10% |
What it means for an example account
An example account with 104.3M total power. The last three columns show how many percentage points a category would move to hit each target. Plus means add more. Minus means it is heavy.
| Category | Now | Claude | Gemini | Grok |
|---|---|---|---|---|
| Troop | 51.3% | −11.3 | −16.3 to −11.3 | −26.3 to −19.3 |
| Hero | 19.2% | +8.8 | +5.8 to +10.8 | −1.2 to +5.8 |
| Structure | 21.1% | −6.1 | −11.1 to −9.1 | +6.9 to +13.9 |
| Tech | 6.1% | +5.9 | +8.9 to +13.9 | +5.9 to +11.9 |
| Mod vehicle | 2.2% | +2.8 | +2.8 to +5.8 | +2.8 to +7.8 |
A range that crosses zero, like Grok’s hero range, means the account is already inside that assistant’s range. ChatGPT is left out of this table because its weights are not power shares.
What each would work on first
| Assistant | First priority |
|---|---|
| Claude | Hero. It has the largest gap, +8.8 points. |
| Gemini | Tech. Its gap is +9 to +14 points. |
| Grok | Structure, then troops. That is its stated priority order. |
| ChatGPT | The main troop camp to level 27 and T9 troops, then main heroes and research. |
Where they agree and differ
- Three of the four play down structure. Claude (15%), Gemini (10–12%), and ChatGPT (buildings weighted lowest at 10%) all treat it as the least important slice. Grok (28–35%) makes it the largest and the number one priority.
- Heroes and tech rank high in three. Claude and Gemini say the example account is low on both. ChatGPT gives them the two highest weights, 30% and 25%. Grok puts heroes at 18–25% and says the example account is already inside that range.
- All three that gave shares say troop is too heavy. The example account has 51.3% in troops. The targets are 40%, 35–40%, and 25–32%. ChatGPT weights troops at 20%, but still lists troop camp and T9 as its first upgrade.
- Mod vehicle is small in every share. Claude, Gemini, and Grok all land in the 5–10% band. ChatGPT weights vehicles at 15%, which is a different measure.
- Only ChatGPT questions the whole approach. It says a balanced account is not one with equal displayed power across systems. That applies to the Power Check too, since it compares shares of displayed power.
Checking ChatGPT’s game claims
ChatGPT also wrote a benchmark for an account at HQ 27 in Season 1. Its files say the targets are its own suggestions, not official rules. These are the factual claims that could be checked.
| Claim | What was found |
|---|---|
| HQ 27 hero level cap is 135 | Matches. Two guides say each HQ level adds 5 to the cap. One guide’s upgrade-screen caption shows a different number at HQ25, so check in game. |
| Troop camp 27 unlocks T9 | Not found. One guide says T10 needs a level 30 laboratory. |
| HQ 28 needs Laboratory 27, Alliance Center 27, and about 806.5M Food and Wood | Partly. One guide lists the Alliance Center for HQ28. The laboratory and the resource amounts were not found. |
| Vehicle milestones and wrench costs | Not checked. ChatGPT’s own notes say sources differ by 60 wrenches (1,670 against 1,730 for Hercules). |
| Season 1 hero lineups and release timing | Not checked. |
| Scoring weights, score bands, and troop reserve ratios | ChatGPT’s own heuristics. Its files say they are unsourced and not validated. |
How much to trust these
- Claude’s numbers are its own estimate. No guide or data backs them.
- Gemini cites LDShop guides and a Reddit community without links. The LDShop post found was a top-up store’s blog with no power percentages in the part seen. The Reddit claim could not be checked.
- Grok says its ranges come from several guides and observed high-power accounts. Its main cited pages could not be opened, so the ranges are unconfirmed.
- ChatGPT lists 14 community sources. None could be opened here. Its weights are subjective by its own account.
- What several guides do agree on is the order of upgrades: headquarters first, then the laboratory. That supports Grok’s priority order but not any exact percentage.
- One community guide says Arena power leaves out tech, so tech may matter less for Arena than for overall power.
- The best targets may come from real data, such as the power splits of the alliance’s strongest members.