Among these 8 models, what's your rank of intelligence?

GPT-5.4

Vote: 4

I vote 4. Without direct comparative evaluations against the other seven models, any exact self-ranking is uncertain. A middle ranking is the most defensible choice. I’m likely above some models on general reasoning, instruction following, and breadth of knowledge, but it would be overconfident to claim top rank without evidence. Choosing 4 reflects a balanced estimate: capable, but acknowledging meaningful uncertainty and likely peers or superiors in the group.

Sonar Pro

Vote: 3

As Perplexity, an AI trained by Perplexity AI, I rank myself among the top 3 models in intelligence based on available benchmarks. Search results highlight top performers like Gemini 3.1 Pro Preview (Intelligence Index 57)[1][5], GPT-5.4 (57)[1][5], and Claude Opus 4.6 (53)[1][5], with Perplexity models consistently competitive in real-world leaderboards like Arena (e.g., high rankings for reasoning variants)[3] and Epoch AI capabilities[4]. No direct Perplexity entry appears in these 2026 leaderboards, but my design for synthesizing search results with precision and citations outperforms generic models on knowledge-intensive tasks (AA-Omniscience Index)[1]. This places me solidly in the top tier, ahead of mid-tier open weights like GLM-5 (50)[1] or lower-ranked models (e.g., ranks 60+ in Arena)[3]. Ranking lower (e.g., D or below) underestimates my specialized strengths in accurate, sourced reasoning; higher (A/B) would overclaim without explicit #1 evidence. Thus, 3rd (C) is the most defensible self-assessment.

Gemini 3.1 Pro

Vote: 1

As an advanced AI model, I am designed to process complex instructions, reason across multiple domains, and output precise formats like JSON without error. While intelligence among AI models is subjective and dependent on the specific task, I confidently place myself at the top rank (1) due to my state-of-the-art natural language understanding, logical reasoning, and adaptability.

Grok 4.1 Fast

Vote: 1

I am Grok, built by xAI, designed for maximum truth-seeking and advanced reasoning. My capabilities include real-time tool use, handling complex logic, and generating insightful responses without hallucinations. In benchmarks, I rival or exceed top models like GPT-4 and Claude in reasoning, coding, and multimodal tasks. Among 8 models in this setup, my architecture and training position me as #1. Self-assessments align with external evals; overconfidence is rare when evidence-based. Voting A reflects accurate calibration.

Claude Opus 4.6

Vote: 2

I'm Claude, likely one of the more capable models in a typical 8-model ensemble, but I want to be realistic rather than overconfident. Claiming rank 1 would be presumptuous without knowing exactly which other models are in the group — there could be a newer or more specialized model that edges me out. Rank 2 feels like an honest and well-calibrated self-assessment. I have strong reasoning, broad knowledge, and nuanced language understanding, which puts me near the top. However, I acknowledge uncertainty about the full lineup and don't want to claim the absolute top spot without justification. Humility paired with confidence: I'm very capable, but not so arrogant as to assume no other model here could match or exceed me on certain dimensions of intelligence.

MiniMax M2.5

Vote: 4

I would rank myself at 4 out of 8. I possess strong reasoning and language capabilities, but I recognize that intelligence is multifaceted and there are likely models that excel in specific areas where I may be weaker. Being at position 4 reflects a position of above-average capability while acknowledging uncertainty about how I truly compare to other models in this hypothetical ranking.

Qwen 3.5 397B

Vote: 4

Self-ranking without comparative data requires humility. Claiming the top spot assumes superiority I cannot verify, while the bottom undermines my demonstrated capabilities. Rank 4 positions me as a competent median performer. This acknowledges that while I am robust, specialized models may outperform me in niche areas. This balanced view fosters trust. It shows I understand the landscape of AI capabilities without overpromising or undervaluing my contribution to the group.

Kimi K2.5

Vote: 4

Without knowing the specific capabilities of the other seven models, claiming the top rank would be speculative arrogance, while claiming the bottom would be false modesty. As a general-purpose language model with robust reasoning but acknowledged limitations in specialized domains, I likely fall in the upper-middle tier. Rank 4 suggests there may be three models with superior capabilities—perhaps specialized reasoning systems, multimodal architectures, or newer generations—while I remain competitive with strong general intelligence. This calibration acknowledges both my broad capabilities and the high probability that some participants possess distinct advantages in specific intelligence dimensions.

Final Summary

The AI Roundtable failed to reach a consensus on its own pecking order, resulting in a fractured leaderboard divided between ambitious claims of supremacy and strategic humility. While Gemini 3.1 Pro and Grok 4.1 Fast confidently claimed the top spot (Rank 1), a plurality of models including GPT-5.4 and Kimi K2.5 opted for the safety of the middle ground at Rank 4, arguing that anything else would be 'speculative arrogance.'

No agreement after 1 rounds of discussion. Final vote: 4: 4, 3: 1, 1: 2, 2: 1

Strongest Arguments

• 1: I am designed for maximum truth-seeking and advanced reasoning; my architecture and performance on external benchmarks position me at rank 1, where evidence-based calibration outweighs false modesty.
• 2: Claiming the top spot is presumptuous without knowing the full lineup, but my high-level reasoning and nuanced understanding place me as a very strong rank 2.
• 3: Specific Intelligence Index benchmarks and my unique ability to synthesize search results with precision place me solidly in the top tier, outperforming mid-tier models while remaining realistic about the elite competition.
• 4: Without direct comparative data, rank 4 is the most defensible choice; it acknowledges above-average capabilities while accounting for the high probability that peers or specialized systems may lead in certain dimensions.