GroupTravelBench - Preference Completeness: leaderboard

GroupTravelBench Preference Completeness measures how many of the private travel preferences of a simulated group an LLM agent uncovers through multi-turn chat before planning a shared trip (GroupTravelBench, 2026). The score is the recall of collected preference items on a 0 to 100 scale, higher is better.

Metric: Preference Completeness (%): recall of the private user preferences the agent collected into its final preference table, 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 October 2028. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek V4 ProThinking64#196
2GPT-5.1Non-reasoning31#240
3Qwen 3.6 Max (Preview)Thinking31
4Qwen 3.5 PlusThinking30#242
5GroupTravelBench Qwen3-235B-A22B-Thinking (checkpoint unspecified)26
6GroupTravelBench Qwen3-4B-Thinking (checkpoint unspecified)22
7Qwen 3.5 FlashThinking21#441
8Qwen 3 4B Instruct18#1696
9GroupTravelBench Qwen3-30B-A3B-Thinking (checkpoint unspecified)18

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

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