Trip+ - Hard Constraints: leaderboard
Trip+ (2026) has tool-using LLM agents plan and revise minute-level travel itineraries over multi-turn conversations with profiled travelers, changing requests and environment disruptions, in a fixed travel sandbox. This board is satisfaction of explicit user requirements, averaged over the plans an agent produces when a plan is expected. Scores are on a 0 to 100 scale; higher is better.
Metric: Hard-constraint satisfaction (0-100), a rate scaled by 100: explicit user requirements (dates, destinations, party size, budget, required lodging, dining and transport) met by rule checks, averaged over the plans that pass the response-mode gate, on multi-turn personalized travel planning with traveler profiles, request changes and environment disruptions, the same OpenAI-compatible function-calling scaffold over a fixed travel sandbox for every model, temperature 0 where supported; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 90.47 | #94 |
| 2 | DeepSeek V4 Pro | 83.75 | #188 |
| 3 | GPT-5.4 | 81.1 | #123 |
| 4 | Seed 2.0 Pro | 80.06 | #211 |
| 5 | Kimi K2.6 | 79.84 | #178 |
| 6 | GLM-5.1 | 79.73 | #179 |
| 7 | Gemma 4 31B | 76.32 | #394 |
| 8 | Qwen 3.5 27B (Non-reasoning) | 72.56 | #359 (Qwen 3.5 27B) |
| 9 | Qwen 3.6 27B (Non-reasoning) | 72.25 | #341 (Qwen 3.6 27B) |
| 10 | Gemini 3 Flash (Preview) | 71.79 | #144 |
| 11 | DeepSeek V3.2 | 71.62 | #406 |
| 12 | Gemma 4 26B A4B | 69.25 | #615 |
| 13 | Hy3-preview | 64.76 | #316 |
| 14 | MiniMax-M2.7 | 63.26 | #518 |
| 15 | GPT-5.4 Mini | 61.26 | #389 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=trip-plus-hard-constraints · How It Works · Data refreshed daily, snapshot 2026-11-13.