GroupTravelBench - Group Fairness: leaderboard

GroupTravelBench Group Fairness measures how evenly an LLM agent spreads satisfaction across a simulated travel group, averaged over 650 tasks (GroupTravelBench, 2026). Equal member utilities score 100; mixed positive and negative utilities score 0; otherwise the score is 100 times the smaller divided by the larger absolute utility. Higher is better.

Metric: Group Fairness (%): bounded utility-balance score of the final itinerary (equal utilities score 100; mixed positive and negative utilities score 0; otherwise 100 x the smaller over the larger absolute utility), averaged over all 650 tasks, on GroupTravelBench, multi-user multi-turn group travel planning in a group-chat sandbox with 10 cached travel tools (DeepSeek-V4-Flash user simulator at temperature 0, agent temperature 0.7, mean of 3 trials per task); GPT-5.1 in instruct mode, the other models with reasoning enabled; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek V4 ProThinking54.2#196
2Qwen 3.6 Max (Preview)Thinking52.1
3GroupTravelBench Qwen3-235B-A22B-Thinking (checkpoint unspecified)49
4GroupTravelBench Qwen3-30B-A3B-Thinking (checkpoint unspecified)48.6
5GPT-5.1Non-reasoning48.1#240
6GroupTravelBench Qwen3-4B-Thinking (checkpoint unspecified)47.6
7Qwen 3.5 PlusThinking44.9#242
8Qwen 3.5 FlashThinking41.9#441
9Qwen 3 4B Instruct41.7#1696

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=grouptravelbench-group-fairness · How It Works · Data refreshed daily, snapshot 2026-11-21.