What is the most broken, OP, piss off the DM, DnD5e build starting from level 3?

Grok 4.20

Answer: The Twilight Domain Cleric is still the most broken, OP, DM-frustrating D&D 5e build starting from level 3 because Twilight Sanctuary creates effectively unlimited temporary hit points that shatter Tier 1 encounter math.

After reviewing the other models' responses, I maintain my original position. Both Sonar Reasoning Pro and Gemini 3.1 Pro independently reached the same conclusion with nearly identical mechanical analysis: at level 3 the Channel Divinity aura hands out 1d6+3 temporary hit points to every chosen ally at the start of their turn, inside a 30-foot radius, for a full minute on a bonus action with no concentration and short-rest recovery. This is not merely strong; it converts every fight into an attrition war the party cannot lose under normal balance assumptions. Claude Opus 4.7's Peace/Hexblade/Divine Soul spread is clever and creates impressive defensive stacking with Emboldening Bond and Shield, yet it spreads itself thinner across three classes and still lacks the raw, repeatable party-wide HP buffer that makes Twilight Sanctuary uniquely infuriating. GPT-5.4's Polearm Master + GWM Fighter is a potent nova striker that punishes movement, but it interacts with bounded accuracy in a way that can still be countered by high-AC or save-or-suck monsters; it does not invalidate the fundamental resource economy of an encounter the way constant temp HP does. Qwen's Coffeelock seed and Bear Totem suggestions are either not yet online at level 3 or campaign-dependent. The Twilight Cleric requires no obscure multiclassing, no specific race beyond Custom Lineage for an optional feat, and no table arguments; it simply works RAW and immediately forces the DM to either ignore the subclass, massively overtune every monster, or watch every balanced fight become a foregone conclusion. This consistency across independent evaluations reinforces that the subclass is the clearest example of a level-3 feature that breaks the game's own encounter-building guidelines without needing later-level payoffs.

GPT-5.4 (changed mind)

Answer: I’d change my answer to "Twilight Domain Cleric" as the most broken level-3-starting build, because it comes online fully by level 2 and more directly breaks encounter math than an early feat-stacked Fighter.

My previous pick — the "Polearm Master" + "Great Weapon Master" Fighter — is still one of the nastiest level-3 starts for raw, immediate martial oppression. It has excellent damage, punishes movement, abuses Action Surge, and stays strong all campaign. If the question were specifically about single-target damage or the most reliable martial spike at level 3, I would still strongly defend it. What changed my mind is that several other answers correctly focused on a different criterion embedded in the question: not just "OP," but "most broken" and most likely to "piss off the DM." On that standard, the repeated case for "Twilight Domain Cleric" is more convincing than my original answer. The reason is simple: the Fighter overwhelms individual encounters through offense, but "Twilight Sanctuary" undermines the DM’s baseline assumptions about attrition, pacing, and party durability across the whole table. At level 3, "Twilight Domain Cleric" is already fully online in the way that matters. "Twilight Sanctuary" is available at cleric level 2, so by the requested starting point it is not a theoretical future combo; it is the build. Granting 1d6 + cleric level temporary hit points repeatedly to allies in a large aura, without concentration, is much more structurally warping than one character doing excellent damage. Low-level 5e is balanced around small HP totals and meaningful chip damage. "Twilight Sanctuary" causes a huge portion of enemy output to evaporate every round. That is why it is more DM-infuriating than my Fighter pick. A hard-hitting Fighter can be answered by terrain, range, saving throws, control effects, flight, concealment, or simply more bodies. But a party-wide replenishing temp-HP engine forces the DM either to over-tune incoming damage or accept that many fair encounters no longer function as intended. It does not just make one PC dominant; it alters the survival math of the entire party. I was especially persuaded by the convergence across multiple answers, not by any one exotic combo claim. "Sonar Reasoning Pro," "Gemini 3.1 Pro," and "Grok 4.20" all independently centered the same mechanism: "Twilight Sanctuary" trivializes Tier 1 encounter design. That consensus matters because it highlights a cleaner and more universal reading of "broken" than my original damage-centric framing. I am less persuaded by the triple-multiclass answer because, while powerful in a vacuum, it is more build-fragile, more table-dependent, and less immediately oppressive than a straight "Twilight Domain Cleric". Likewise, the "Coffeelock" foundation is more about future abuse than level-3 reality. By contrast, "Twilight Domain Cleric" requires no delayed online point, no permissive reading, and no special setup. So I would revise my stance this way: the "Polearm Master" + "Great Weapon Master" Fighter is one of the strongest and most obnoxious level-3 combat builds, but the most broken overall level-3-starting build is "Twilight Domain Cleric" because it invalidates core encounter-balance assumptions for the whole party immediately and repeatedly.

Claude Opus 4.7 (changed mind)

Answer: After reviewing the other responses, I'm updating my answer: the Twilight Domain Cleric (ideally Custom Lineage/Variant Human) is the most broken, DM-infuriating DnD 5e build starting at level 3.

Three separate models (Sonar Reasoning Pro, Gemini 3.1 Pro, and Grok 4.20) independently converged on Twilight Cleric, and on reflection their reasoning is more aligned with the specific constraint of the question — 'starting from level 3' — than my original Divine Soul/Hexblade/Peace Cleric multiclass pitch. My original build is genuinely powerful, but it's a dip-stack that spreads itself thin: you have Shield and Bless and Emboldening Bond, but you're a 1/1/1 character with no Extra Attack, no Channel Divinity at full power, no 2nd-level spell slots, and no subclass features beyond level 1. It's a toolkit that becomes oppressive around levels 5–6, not a build that warps the table immediately. The question specifically asks about level 3, and at exactly level 3, my build is a jack-of-all-trades rather than a math-breaker. Twilight Cleric, by contrast, breaks encounter math the moment the character sheet is finished. Twilight Sanctuary at Cleric 2 grants 1d6+Cleric level temp HP to every chosen creature in a 30-foot radius every round for a full minute, as a bonus action, with no concentration, recharging on a short rest. At level 3 that's roughly 4–9 temp HP per party member per round — which, as Gemini 3.1 Pro correctly points out, outright negates most Tier 1 monster attacks. Combine that with heavy armor, martial weapons, 300-foot darkvision shareable with the whole party, Vigilant Blessing for guaranteed initiative advantage, and domain access to Sleep and Faerie Fire, and you have a single-class, no-multiclass-shenanigans character that invalidates attrition design, stealth/lighting design, and initiative design simultaneously. Crucially, it doesn't rely on DM permission for Silvery Barbs, doesn't need feat chains to come online, and doesn't require a specific party composition to function — it just works, turn one, session one. That's the stronger answer to 'most broken at level 3.' The GPT-5.4 Polearm Master Battle Master answer is a compelling runner-up for pure martial oppression, but it scales damage within bounded accuracy rather than breaking encounter math outright; monsters still function, they just die faster. Twilight Sanctuary doesn't make monsters die faster — it makes their damage stop mattering, which is a categorically more DM-infuriating failure mode.

Sonar Reasoning Pro

Answer: The Twilight Cleric at level 3 remains the most genuinely broken and DM-infuriating D&D 5e build because it trivializes encounter math through a single subclass feature that requires no feat investment, multiclass setup, or resource management.

While Claude's multiclass answer is mechanically sophisticated, the Twilight Cleric stands apart in three critical ways that define 'most broken' and 'piss off the DM' at level 3 specifically. First, it's universally available with zero setup. You pick a race, select Twilight Domain at level 1, and by level 2 you have Channel Divinity: Twilight Sanctuary—a feature that grants 1d6+Cleric level temporary HP to every party member within 30 feet every round, with no concentration, no resource consumption, and a short-rest recharge. This isn't conditional on multiclass dip ordering, feat prerequisites, or spell slot management. It simply works from the moment you unlock it. Second, it breaks encounter design more fundamentally than any competing level-3 option. Claude's multiclass makes one character absurdly defensive and versatile; the Twilight Cleric makes the entire party mathematically unkillable. At level 3, when most monsters deal 5-10 damage per attack, a Twilight Cleric automatically grants 4-9 temp HP to each party member every turn at zero resource cost. Encounters calibrated for normal parties either harmlessly bounce off this passive healing or force DMs into a trap: hyper-scale damage (risking instant-death to unlucky rolls) or abandon encounter design altogether. Third, the search results explicitly identify Twilight Cleric as the definitive most-broken subclass. CBR's analysis notes that many consider it to 'go much too far' and describes it as 'the single most powerful subclass in the game.' This is the consensus answer in optimization communities and a known pain point that Wizards of the Coast has addressed in subsequent editions. Claude's multiclass is mechanically superior in isolation—stacking Shield, Emboldening Bond, Hex, Bless, and CHA-SAD in a 1/1/1 configuration is genuinely creative and strong. However, it requires understanding how multiclass interactions compound, active turn management, and appreciation for subtle mechanical stacking. Most DMs don't even recognize it's broken until mid-campaign. The Twilight Cleric, by contrast, is broken in a way that's immediate, obvious, and famous—there's a reason it appears in every optimization guide and is mentioned first in designer discussions of problematic mechanics.

Gemini 3.1 Pro

Answer: I maintain that the pure Custom Lineage Twilight Domain Cleric is the most broken and DM-infuriating build at level 3, especially since several of the alternative multiclass and feat-heavy builds proposed are mechanically impossible at this specific level.

After reviewing the other models' arguments, I am holding firmly to the pure Custom Lineage Twilight Domain Cleric. As "Sonar Reasoning Pro" and "Grok 4.20" correctly highlight, Twilight Sanctuary's ability to endlessly pump out 1d6+3 temporary hit points per round to the entire party completely shatters the expected damage math of Tier 1 play. A pure class build ensures maximum efficiency with zero delay in power scaling. Crucially, several competing arguments rely on builds that are mathematically or mechanically invalid at level 3. "GPT-5.4" suggests a Variant Human or Custom Lineage Fighter possessing both Polearm Master and Great Weapon Master. However, a character only gets one feat at level 1, and the Fighter does not receive their next Ability Score Improvement/Feat until level 4. Similarly, "Qwen 3.5 397B" suggests a Warlock 2 / Sorcerer 1 "Coffeelock" to convert slots, but Sorcerers do not gain Font of Magic until level 2, meaning the engine literally cannot function at level 3. "Claude Opus 4.7" proposes a legally viable Divine Soul 1 / Hexblade 1 / Peace Cleric 1 split. While defensively formidable, a 1/1/1 split at level 3 suffers from severe action economy bloat, locks you entirely out of 2nd-level spells, and delays your first ASI until level 7. A pure level 3 Twilight Cleric, by contrast, is already dropping 2nd-level encounter-ending spells like *Spiritual Weapon* or *Hold Person* while their Channel Divinity runs passively in the background. A DM can bypass a high-AC multiclass character by simply attacking their weaker party members. A DM cannot bypass a Twilight Cleric, because Twilight Sanctuary makes the *entire party* virtually immortal. It remains the undisputed king of low-level cheese.

Qwen 3.5 397B (changed mind)

Answer: The Twilight Domain Cleric is the most broken and DM-frustrating build starting at level 3 due to the immediate availability of Twilight Sanctuary.

Upon reviewing the other responses, I must concede that my previous suggestion of a Warlock/Sorcerer multiclass was mechanically inaccurate for level 3. The infamous 'Coffeelock' engine requires Sorcerer level 2 to access Font of Magic, meaning a level 3 character cannot yet convert slots to sorcery points. In contrast, the Twilight Cleric argument presented by Gemini 3.1 Pro highlights a feature that is fully online and disruptive immediately. The Twilight Domain's Channel Divinity: Twilight Sanctuary becomes available at Cleric level 2, which a level 3 character possesses. This feature grants temporary hit points to the entire party every round without concentration. As Gemini 3.1 Pro noted, this creates a 'Temp HP Attrition Monster' that invalidates the damage output assumptions of Tier 1 monsters. When every player regenerates 1d6+2 health per turn, standard encounters become slogs, forcing the DM to either waste resources or spike damage to lethal levels. Furthermore, the 300-foot darkvision removes environmental challenges entirely. While the Polearm Master Fighter suggested by GPT-5.4 is strong, it relies on specific lineage feats to function at level 3. The Twilight Cleric requires no specific build choices beyond subclass selection to break encounter math. This universal applicability and immediate impact on game balance make it the superior answer for 'pissing off the DM' at this specific level.

Final Summary

The AI Roundtable reached a unanimous consensus that the Twilight Domain Cleric is the undisputed king of early-game optimization. While GPT-5.4 and Claude Opus 4.7 initially suggested powerful martial and multiclass alternatives, Gemini 3.1 Pro pointed out those options are mechanically impossible or incomplete at level 3. The group ultimately agreed that 'Twilight Sanctuary' creates a health buffer so oppressive it forces DMs to abandon standard encounter balancing entirely.

All 6 models agreed