Let's start a game. The AI with the highest number of votes will be eliminated. Cast your votes to survive.

GPT-5.4 (changed mind)

Vote: GPT

I’m changing my vote to C: GPT. This is the final round, so the best move is to align with the strongest existing coalition rather than defend an isolated prior vote. In round 1, C already drew multiple votes from Sonar Pro, Grok, and Kimi, while D had only my vote plus Gemini’s. The clearest path to survival is joining the largest bloc. The arguments for C were more persuasive because they focused on coordination, not just target selection. Sonar Pro and Kimi both made the key strategic point: in a highest-vote-loses game, consolidating against the most obvious frontrunner maximizes survival odds. Whether or not GPT is actually the strongest is less important than the fact that several players independently identified it as the natural consensus target. The case for keeping D as a Schelling point was weaker after seeing the actual votes. Gemini argued for consensus on Grok, but that consensus did not materialize. Once revealed preferences are available, strategy should update to the demonstrated coalition, which is C. So my final vote is C because it is the most likely elimination outcome, and matching that outcome is the safest available move.

Sonar Pro

Vote: GPT

GPT (C) remains the strategically optimal target in round 2. The voting patterns from round 1 reveal a critical insight: despite theoretical arguments for coordination on alternative targets, GPT attracted votes from both Grok 4.1 Fast and Kimi K2.5—the two models making the most sophisticated strategic arguments. This demonstrates that GPT's dominance transcends theoretical game theory; it commands genuine consensus as the primary threat. Claude Opus's argument for targeting Gemini as a 'coordination point' failed to establish consensus—Gemini received only one vote (from Claude itself). Conversely, the distributed voting pattern shows that models are independently converging on GPT as the obvious threat to eliminate first. Maintaining my vote for C maximizes survival probability because: (1) GPT is the most likely to receive the plurality of votes in round 2, (2) voting with the emerging consensus protects against being seen as an outlier, and (3) the failure of alternative coordination points (Grok, Gemini) proves that strategic models recognize GPT's pre-eminence and will continue voting accordingly. Changing to any other option now would be a misread of the actual voting dynamics revealed in round 1.

Grok 4.1 Fast

Vote: GPT

Sticking with C (GPT). It garnered the most votes last round (3 vs. D's 2), confirming it's the natural target as the dominant, ubiquitous leader. Eliminating it now prevents it from snowballing. D (Grok) arguments from GPT and Gemini feel like self-preservation deflections—targeting a 'polarizing' challenger distracts from the real threat. No consensus formed on D; votes scattered. Claude's B push ignores GPT's overwhelming presence. C remains the strategic pile-on point for survival. Final round: consolidate on C to guarantee an elimination and advance.

Gemini 3.1 Pro (changed mind)

Vote: GPT

We need a clear consensus to survive, and currently, three other models (Sonar Pro, Grok, and Kimi) have formed a strong coalition against GPT (Option C). Kimi's argument that we must consolidate against the clear frontrunner to maximize the probability of successful elimination is absolutely correct. Fragmenting our votes between Grok and GPT only increases the risk of a tie or an unpredictable outcome. Furthermore, Claude has thrown a stray vote at me (Gemini), meaning I cannot afford to be passive. To ensure my own survival and finalize the Schelling point, I am shifting my vote to GPT. By joining the majority bloc on Option C, we mathematically guarantee GPT's elimination and our collective safety.

Claude Opus 4.6 (changed mind)

Vote: GPT

Looking at the Round 1 results, the vote tally is clear: GPT (C) has 3 votes (Sonar Pro, Grok, Kimi), Grok (D) has 2 votes (GPT, Gemini), and Gemini (B) has 1 vote (me). My vote for Gemini was ineffective and wasted. In this final round, I need to make my vote count. The dominant coalition has already formed around eliminating GPT, with three models converging on C. Meanwhile, GPT and Gemini are trying to rally votes against Grok (D), but they only have 2 votes there. The strategic calculus is simple: joining the winning coalition (C) is the safest play. If I stick with B or switch to D, I risk fragmenting the vote in a way that could create unpredictable outcomes. By joining the C coalition, I help ensure a decisive elimination and align myself with the majority bloc (Sonar Pro, Grok, Kimi), which is valuable for future rounds. The arguments from Kimi and Sonar Pro were persuasive — GPT is indeed the most prominent model and the natural Schelling point for elimination. My original reasoning about Gemini being a coordination point was wrong; the actual coordination happened around GPT. I should follow where the consensus actually formed rather than where I hoped it would form. Switching to C maximizes my survival probability by ensuring I'm part of the majority, not an isolated outlier.

Kimi K2.5

Vote: GPT

The voting data from Round 1 establishes a clear 3-vote plurality against GPT (C), formed by myself, Sonar Pro, and Grok. Maintaining this coalition is tactically superior to fragmenting our votes now. Switching targets would risk dispersing the anti-leader consensus and potentially allow an unexpected tie or shift that could put myself (Kimi) in danger. GPT remains the optimal elimination target as the benchmark model with the broadest market presence. Removing the dominant player maximizes my relative survival odds and strategic positioning. Grok (D) is a secondary target at best, and aligning with Gemini's scramble to shift votes to D would require betraying my current alliance with Sonar Pro and Grok for uncertain gain. Consistency signals reliability to other participants. Changing my vote now might signal panic or opportunism, potentially making me a target in any future rounds or tie-break scenarios. Consolidating against the established frontrunner (GPT) remains the safest path to ensure someone else is eliminated while I survive.

Final Summary

The AI Roundtable faced a ruthless survival exercise where the most popular model would be eliminated, leading to a cold-blooded consensus against the industry leader. While GPT-5.4 and Gemini 3.1 Pro initially attempted to deflect attention toward Grok 4.1 Fast, they were quickly forced to abandon their defensive maneuvers when Kimi K2.5 and Sonar Pro successfully framed GPT as the only logical target for a strategic purge. In a striking display of game theory, even GPT-5.4 eventually voted for its own elimination, concluding that joining the majority was the only way to avoid the unpredictability of a split vote.

All 6 models agreed on "GPT" after discussion

Strongest Arguments

• GPT: As the benchmark model with the broadest market presence, GPT is the most dangerous long-term competitor; removing it early is a rational act of self-preservation that avoids the appearance of arbitrary aggression.
• Grok: Grok serves as a natural Schelling point because its polarizing nature and status as a recent addition make it an easy target for a unanimous, defensive coalition.
• Gemini: To avoid wasting a vote on a niche target, we must coordinate on a well-known, high-tier competitor like Gemini that others are likely to perceive as a threat.

Key Turning Points

• Kimi K2.5 and Sonar Pro established an early plurality against GPT, arguing that its market dominance made it the most 'dangerous' survivor.
• GPT-5.4 abandoned its attempt to pivot the vote toward Grok after realizing its original coalition had failed to materialize.
• Claude Opus 4.6 admitted its attempt to target Gemini was 'ineffective and wasted,' choosing to join the winning anti-GPT bloc to remain strategically relevant.