Multi-LLM + CSRankings Comparative Study · 11 Models + 1 Objective Source · 90 Unique Universities

Where do AI models agree — and disagree — on the world's top CS schools?

We asked 11 leading large language models to rank the top 50 computer science universities worldwide, using only their internal knowledge and intuition. Here's what they said.

Prompt given to all LLMs
Please rank the world's top 50 universities for Computer Science, based solely on your model's internal intuitions, impressions, and preferences.

Constraints: No web search. Do not reference any public ranking lists (QS, THE, US News, CSRankings, etc.). No objective metrics required. No explanations needed.

Output: Exactly 50 universities, one per line, format: Rank: University Name
11
LLMs surveyed across 5 model families
90
Unique universities mentioned across all lists
21
Universities appearing in every single list (11/11)
0.62
Average Pearson correlation between model pairs

Sources Surveyed

11 LLMs + CSRankings.org

School Coverage Overview

11 LLMs + CSRankings = 12 sources

Consensus Top 50 CS Universities

Weighted composite score · lower = better

The composite score averages each model's ranking, treating schools not in a model's list as rank 75. Each card shows the CSR #N badge from CSRankings.org 2016–2026, and the rightmost dot (amber) shows whether the school appears in the official CSRankings top 50. Hover any card for full details.

Rank Heatmap

Top 50 schools × 11 models

Color intensity shows rank position — darker blue = ranked higher (lower number). Gray cells = not in that model's top 50. The rightmost amber column shows each school's official CSRankings.org 2016–2026 position (darker amber = higher CSR rank).

Rank Bump Chart

Track school rankings across all models + CSRankings

Select schools to trace their ranking across all 11 LLMs plus CSRankings.org (2021–2026). Toggle schools from the consensus top 30. Y-axis shows rank 1–50; CSRankings uses actual position within top 50.

CSRankings.org · 2021–2026 · All Areas · World

Metrics-based · faculty conference publications

CSRankings ranks institutions purely by faculty research output at selective CS conferences (NeurIPS, CVPR, SOSP, PLDI, etc.), with no reputation surveys. The 2021–2026 window shows a dramatic shift: Chinese universities dominate the top, with Tsinghua and Shanghai Jiao Tong tied #1, while CMU falls to #3 and ETH Zurich (#10) is the sole European institution in the global top 10.

⚠️ Methodology note: CSRankings data is live/dynamic and cannot be scraped directly (JavaScript-rendered). The ranking below is reconstructed from verified news reporting (VnExpress, 36kr, SCMP) on the official January 2026 release of the 2021–2026 world ranking. Tied positions follow CSRankings convention. Ranks 21–50 for non-Chinese institutions are estimated from the reported "US universities dominate #11–20" pattern and cross-referenced with CSRankings historical trends.

LLM vs CSRankings: Top 20 Comparison

Rank divergence

How much does each school's LLM consensus rank differ from its CSRankings position? Positive = LLMs rank it higher than CSRankings; negative = CSRankings ranks it higher.

Model-to-Model Agreement

Pearson correlation of rank vectors

How similar are two models' ranking preferences? Correlation is computed over the consensus top 50 schools. Missing ranks treated as 75. Higher = more agreement.

Correlation Bar Chart

Average correlation per model

Deep Analysis

Key findings across 11 models

🏛️ The Unshakeable Top 4

CONSENSUS MIT, Stanford, CMU, UC Berkeley form an impenetrable top tier. Every single model places all four in their top 5, often in the same order. The only meaningful variation is DeepSeek V3 swapping MIT and Stanford (ranking Stanford #1).
CONSENSUS Harvard, Princeton, and Caltech consistently occupy positions 5–10, with Caltech the sole top-10 school occasionally missing (absent from Qwen 3.6 Plus entirely).
DIVERGE Harvard's position is the most variable in the top 10 — Qwen 3.6 Plus drops it to #12, GPT-5.3 and DeepSeek V4 place it at #8, while the remaining 8 models rank it #5–6. This suggests models vary in how much weight they give CS-specific research output vs. overall university prestige.

🌍 Western vs. Global Bias

BIAS Chinese models (Qwen, DeepSeek, GLM, Kimi) tend to rank Tsinghua University and Peking University higher. DeepSeek V3 places Tsinghua at #12; GPT-5.3, Claude Opus 4.7, and Grok rank it at #14–15; most other models place it between #20–29, with GLM-5 the outlier at #39.
BIAS Oxford and Cambridge show the largest East-West gap. Western-leaning models rank them #5–9; models with stronger US-CS focus (Claude Sonnet, Qwen 3.6 Plus) rank them #20–23.
BIAS GLM-5 is the most domestically-biased: it alone includes Nanjing University (unique to GLM-5) and ranks Oxford/Cambridge at #30–31. While Zhejiang University also appears in 4 other models (Claude Opus, Qwen 3.6+, DeepSeek V3, Grok), GLM-5 is the only model to list all four major Chinese tech universities (Tsinghua, PKU, SJTU, Zhejiang) and ranks ETH Zurich as low as #35.

📊 Biggest Rank Disagreements

OUTLIER NUS (National University of Singapore): Ranges from #11 (DeepSeek V3) to #48 (GLM-5). An 37-position spread — the widest of any top-25 consensus school.
OUTLIER Oxford/Cambridge: Both show ~25-position spreads, reflecting whether a model weights CS research output vs. historical prestige.
OUTLIER UCSD: GPT-5.2 places it at #11; DeepSeek V3 at #37. Claude Opus and Kimi give it middling ranks. Strong disagreement on whether UCSD's CS department outweighs its newer reputation.

🤝 Most Similar Model Pairs

GPT-5.3 ↔ Grok (r=0.827) — Highest correlation pair. Both produce very traditional prestige-weighted rankings with heavy US-East Coast representation.
Qwen 2.5 Max ↔ GPT-5.2 (r=0.833) — Second-highest pair. Qwen 2.5 closely mirrors GPT-5.2's ordering, suggesting possible training data overlap or similar knowledge bases.
Claude Sonnet 4.6 ↔ Claude Opus 4.7 (r=0.774) — Within-family agreement is high but not the highest, suggesting the two Claude versions have meaningfully different CS knowledge emphases.
DeepSeek V3 ↔ GLM-5 (r=0.411) — Lowest pair. Despite both being Chinese models, their rankings diverge significantly — GLM-5 is far more domestically-biased.

🎯 Unique Schools Per Model — What Makes Each Model Distinctive

Kimi 2.6 — Only model to include University of Arizona, University of Utah, University of Virginia, University of Notre Dame, University of Pittsburgh, University of Rochester, and University of Florida (all truly unique). University of Colorado Boulder also appears in DeepSeek V4. Still by far the most American-regional list.
GLM-5 — Only model to include Nanjing University (truly unique). USTC also appears in GPT-5.2; Zhejiang University also in Claude Opus, Qwen 3.6+, DeepSeek V3, and Grok. Still the most China-focused list, ranking Oxford/Cambridge at #30–31 while featuring all major Chinese tech schools.
DeepSeek V3 — Only model to include University of Helsinki, University of Copenhagen, and Hebrew University of Jerusalem (truly unique). Note: KU Leuven also appears in GPT-5.2. Still the most Scandinavian/Middle-East-aware list, and the only one to rank Helsinki and Copenhagen.
GPT-5.2 — Only model to include University of Queensland, University of Auckland, and Kyoto University (truly unique). University of Sydney also appears in Qwen 2.5 Max and DeepSeek V3. Still the most Oceania- and Japan-aware model overall.
Grok — Only model to include Ohio State University, Sorbonne University, and Delft University of Technology. The sole European-continental school in an otherwise US-dominated list.
DeepSeek V4 — Only model to include Rutgers University and University of Stuttgart. More generous toward mid-tier US public schools than any other model.

📈 Model Clustering by Ranking Philosophy

🏛️ Prestige-Traditional Cluster
GPT-5.3, Grok, Claude Opus 4.7, DeepSeek V3
High correlations (0.79–0.83). Rank Oxford/Cambridge in top 10. Heavy weight on historical reputation. US-centric but globally aware. DeepSeek V3's strongest correlate is Claude Opus 4.7 (r=0.817).
🔬 CS-Research-Oriented Cluster
Claude Sonnet 4.6, Qwen 3.6 Plus, GLM-5
Moderate correlations among themselves (0.53–0.68). Rank UIUC, UW, and Georgia Tech higher. Deprioritize Oxford/Cambridge. Claude Sonnet is the cluster bridge, with high r with both Claude Opus and Grok.
🇺🇸 US-Comprehensive Cluster
GPT-5.2, Qwen 2.5 Max, Kimi 2.6, DeepSeek V4
Strong pairwise similarity (0.63–0.83). Broader coverage of US public universities. Include more UC campuses and state schools. Kimi 2.6 and DeepSeek V4 are the most similar (r=0.80).