cs.CLAug 14, 2026
SaveA Survey of Large Models in Sports
Organizations: Renmin University of China · Sichuan University · Beijing University of Posts and Telecommunications
Abstract
Sports have witnessed growing global enthusiasm in recent years, serving as a vital force for physical health, cultural exchange, social connection, and economic growth. The rapid advancement of large models, particularly (multimodal) large language models (M)LLMs, has demonstrated transformative potential to reshape sports understanding, analysis, and interaction across diverse domains. This paper presents a comprehensive survey of large models in sports, including (i) an overview of tasks and applications across different participant groups; (ii) a detailed analysis of sports-related datasets and benchmarks; and (iii) a critical discussion of current challenges and future directions. Our goal is to establish a foundation for advancing research and practical development of large-model-driven sports intelligence. An open-source GitHub repository is maintained at: https://github.com/Road2Redemption/Awesome_Large_Models_In_Sports1.
Figures & tables
Figure 1: Large models have shown powerful applications across 6 sports stakeholder groups: athletes and trainers, coaches and educators, referees, researchers, fans and social media, and the sports industry, enabling diverse tasks.
| Dataset | Model | Architecture | Paradigm | Metric | Performance |
| Exercise and Training Plans | |||||
| YourSkatingCoach ( 2024c ) | MAAIG ( 2023 ) | T5 (pretrained) ( 2020 ) | fine-tuning | BLEU-4 | 0.27 |
| Transformer ( 2017 ) | vanilla Transformer | train from scratch | 0.04 | ||
| Action Spotting and Recognition | |||||
| SoccerNet-v2 ( 2021 ) | Soccer-CLIP ( 2025 ) | CLIP ( 2021 ) | fine-tuning | t-AmAP | 75.7 |
| COMEDIAN ( 2024 ) | spatiotemporal Transformer | train from scratch | 73.1 | ||
Table 1: Quantitative comparison of modeling paradigms across 4 representative sports tasks.
Figure 3: Data analysis from 3 different perspectives. Sports and tasks abbreviations are listed in Table 6 .
| Benchmark | Sports | Modal | # Video | # QA |
| BIG-bench-SU ( 2023 ) | SC, BK, etc. | text | - | 986 |
| SportQA ( 2024a ) | TN, AF, etc. | text | - | 70592 |
| SPORTU ( 2025 ) | BB, IH, etc. | video, text | 1701 | 12948 |
| Sports-3K-QA ( 2025b ) | 49 Sports | video, text | 412 | 1174 |
| FSBench ( 2025a ) | FS | video, text | 783 | 4000 |
| FBBench ( 2025b ) | BM | video, text | 2563 | 2563 |
Table 2: Overview of specialized sports understanding benchmarks related to large models. Sports abbreviations are listed in Table 6 .
Figure 4: Papers on large models in sports over the years (data for 2025 is up to July).
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| Task | Dataset | Sports | Modal | Method | Related Large Model | Performance | Link |
| Athletes and Trainers | |||||||
| PLA | YourSkatingCoach ( 2024c ) | Figure Skating | V, T | MAAIG ( 2023 ) | T5 ( 2020 ) | 22.08 (METEOR) | ✗ |
| PACE ( 2022 ) | Fitness | T | Hegde et al. (2024) | LaMDA ( 2022 ) | 3.78 ± 1.00 / 5.00 (Likert) | ✓ | |
| NSCA-CSCS ( 2025 ) | Fitness | T | PH-LLM ( 2024 ) | Gemini Ultra 1.0 ( 2023 ) | 88.00 (Acc) | ✗ | |
| T3Set ( 2025a ) | Table Tennis | V, M, T | SenseCoach ( 2025a ) | Llama 3.3-70B ( 2024 ) | 51.64 (P@6-S1L) | ✓ | |
| SCD ( 2025 ) | Soccer | T | Han (2025) | BERT ( 2019 ) | 85.64 (BERTScore) | ✗ | |
Table 3: Summary of task-specific sports datasets related to large models, including athletes and trainers, coaches and educators, and referees. Task: PLA: exercise and training plans, INJ: sports injury and rehabilitation, PSY: sports psychology and behavior, ACT: action spotting and recognition, AQA: sports action quality assessment, TAC: sports tactics and strategies, PRD: game and player performance prediction, REF: sports refereeing. Sports: SC: soccer, BK: basketball, CL: sports climbing, AF: American football, TN: tennis, BM: badminton, TT: table tennis, GY: gymnastics, DV: diving. Modal: V: video, I: image, A: audio, S: skeleton data, M: IMU data, T: text.
| Task | Dataset | Sports | Modal | Method | Related Large Model | Performance | Link |
| Fans and Social Media | |||||||
| CMT | Custom Dataset ( 2022 ) | Soccer | V, A, T | Gautam et al. (2022) | GPT-3 ( 2020 ) | 0.31 (ROUGE-L) | ✓ |
| SN-Caption-test-align ( 2024 ) | Soccer | V, T | MatchVoice ( 2024 ) | Llama 3 ( 2024 ) | 42.00 (CIDEr) | ✓ | |
| LoL19 ( 2024 ) | Esports | T | Wang and Yoshinaga (2024) | Llama 2 13B ( 2023b ) | -4.61 (BARTScore) | ✓ | |
| Custom Dataset ( 2024b ) | Soccer | V, T | AiCommentator ( 2024b ) | GPT-3.5-turbo ( 2022 ) | 0.56 (Cohen’s d) | ✗ | |
| CommentarySet ( 2024 ) | TF,SC,BK,GY,TT,TN | V, T | Ge et al. (2024) | InternVL-Chat-2 ( 2024d ) | 5.44 (SCORES) | ✗ | |
Table 4: Summary of task-specific sports datasets related to large models, including fans and social media, and the sports industry. Task: CMT: sports commentary generation, HLG: sports highlight generation, NSG: sports news generation, NAR: sports narratives and storytelling, OPI: public opinion analysis in sports, MOD: sports models and systems, MAN: sports management, TSC: sports talent scouting. Sports: TF: track and field, SC: soccer, BK: basketball, TN: tennis, TT: table tennis, GY: gymnastics, DV: diving. Modal: V: video, I: image, A: audio, T: text.
| Dataset | Sports | Source | Annotation | Benchmark | Modal | # Video | Avg. Length | # QA | Link |
| Specialized Sports Understanding Datasets | |||||||||
| QASports ( 2023 ) | SC, BK, AF | Fandom | auto | ✗ | T | - | - | 1500K | ✓ |
| BIG-bench-SU ( 2023 ) | SC, BK, AF, BB, IH | program | crowd | ✓ | T | - | - | 986 | ✓ |
| SportQA-Level-1 ( 2024a ) | - | existing (dataset) | manual | ✓ | T | - | - | 21385 | ✓ |
| SportQA-Level-2 ( 2024a ) | 35 Sports | Wikipedia | expert | ✓ | T | - | - | 45685 | |
| SportQA-Level-3 ( 2024a ) | SC, BK, TN, AF, TT, VB | expertise | expert | ✓ | T | - | - | 3522 | |
Table 5: Summary of large-model-related datasets specifically for sports understanding and general video understanding with sports content. Sports: SC: soccer, BK: basketball, BM: badminton, TN: tennis, GY: gymnastics, FS: figure skating, AF: American football, BB: baseball, CR: cricket, TT: table tennis, VB: volleyball, DV: diving, IH: ice hockey, CL: sports climbing, SW: swimming. Source: syn.: synthesis. Modal: V: video, A: audio, T: text.
| Abbr. | Sports | Abbr. | Tasks |
| AF | American Football | ACT | Action Spotting and Recognition |
| BB | Baseball | AQA | Sports Action Quality Assessment |
| BK | Basketball | CMT | Sports Commentary Generation |
| BM | Badminton | EDU | Sports Education |
| BX | Boxing | HLG | Sports Highlight Generation |
| CL | Sports Climbing | INJ | Sports Injury and Rehabilitation |
Table 6: Abbreviations for sports and tasks mentioned in this paper (sorted alphabetically by abbreviation).