| % refs.bib -- ThakiCloud AI Research |
| % All arXiv entries verified against the arXiv API on 2026-09-23. |
| % Own-prior entries verified via the Hugging Face dataset API on 2026-09-23. |
| % BFCL web entry verified (HTTP 200) on 2026-09-23. |
|
|
| % ---------------- Own prior work (ThakiCloud) ---------------- |
|
|
| @techreport{thaki2026ensamblerouter, |
| title = {Cross-Tier Model Ensemble Voting for Skill Routing: A Cost--Quality Study Beyond Single-Signal Retrieval Fusion}, |
| author = {{ThakiCloud AI Research (Hyojung Han)}}, |
| year = {2026}, |
| month = aug, |
| institution = {ThakiCloud}, |
| url = {https://thakicloud.com/tech-blog/ko/research/ensemble-tier-voting-skill-router/}, |
| note = {Dataset: \url{https://huggingface.co/datasets/thaki-AI/daily-paper-2026-08-06-ensemble-tier-voting-skill-router}} |
| } |
|
|
| @techreport{thaki2026toolcallcliff, |
| title = {The Tool-Call Cliff: Measuring the Accuracy Decay of Agentic Structured Output Under Low-Bit Quantization in Self-Hosted H200 Serving}, |
| author = {{ThakiCloud AI Research (Hyojung Han)}}, |
| year = {2026}, |
| month = sep, |
| institution = {ThakiCloud}, |
| url = {https://thakicloud.com/tech-blog/ko/research/tool-call-quantization-cliff/}, |
| note = {Dataset: \url{https://huggingface.co/datasets/thaki-AI/daily-paper-2026-09-17-tool-call-quantization-cliff}} |
| } |
|
|
| @techreport{thaki2026verbositytax, |
| title = {The Reasoning Verbosity Tax: Per-Turn Cost--Quality Frontiers of Explicit vs.\ Compact Chain-of-Thought in Self-Hosted Agentic LLMs on H200}, |
| author = {{ThakiCloud AI Research (Hyojung Han)}}, |
| year = {2026}, |
| month = sep, |
| institution = {ThakiCloud}, |
| url = {https://thakicloud.com/tech-blog/ko/research/reasoning-verbosity-tax-agentic-cost/}, |
| note = {Dataset: \url{https://huggingface.co/datasets/thaki-AI/daily-paper-2026-09-11-reasoning-verbosity-tax-agentic-cost}} |
| } |
|
|
| @techreport{thaki2026manifestorder, |
| title = {The Manifest Order: Measuring KV-Cache Prefix Reuse and Token Cost Savings from Skill-Manifest and Tool-Schema Ordering in Unattended Agentic Loops on H200}, |
| author = {{ThakiCloud AI Research (Hyojung Han)}}, |
| year = {2026}, |
| month = sep, |
| institution = {ThakiCloud}, |
| url = {https://thakicloud.com/tech-blog/ko/research/agentic-prefix-reuse-manifest-ordering/}, |
| note = {Dataset: \url{https://huggingface.co/datasets/thaki-AI/daily-paper-2026-09-20-agentic-prefix-reuse-manifest-ordering}} |
| } |
|
|
| % ---------------- External: post-2026-03 (verified 2026-09-23) ---------------- |
|
|
| @misc{arxiv260528840, |
| title = {How Consistent Are {LLM} Agents? Measuring Behavioral Reproducibility in Multi-Step Tool-Calling Pipelines}, |
| author = {Yagubyan, Abel}, |
| year = {2026}, |
| month = apr, |
| eprint = {2605.28840}, |
| url = {https://arxiv.org/abs/2605.28840} |
| } |
|
|
| @misc{arxiv260508680, |
| title = {Semantic Voting: Execution-Grounded Consensus for {LLM} Code Generation}, |
| author = {Jiang, Shan and Yi, Zijian and Zhu, Chenguang}, |
| year = {2026}, |
| month = may, |
| eprint = {2605.08680}, |
| url = {https://arxiv.org/abs/2605.08680} |
| } |
|
|
| @misc{arxiv260805643, |
| title = {Refining Over Resampling: Test-Time Self-Correction for {LLM} Reasoning}, |
| author = {Bilal, Ahsan and Mohsin, Muhammad Ahmed and Umer, Muhammad and Trigg, Lena}, |
| year = {2026}, |
| month = aug, |
| eprint = {2608.05643}, |
| url = {https://arxiv.org/abs/2608.05643} |
| } |
|
|
| @misc{arxiv260806370, |
| title = {The Bitter Lesson of Tool Calling}, |
| author = {Patel, Ishan and Sen, Sahil and Lumer, Elias and Subbiah, Vamse Kumar}, |
| year = {2026}, |
| month = aug, |
| eprint = {2608.06370}, |
| url = {https://arxiv.org/abs/2608.06370} |
| } |
|
|
| @misc{arxiv260327905, |
| title = {ATLAS-RTC: Closing the Loop on {LLM} Agent Output with Token-Level Runtime Control}, |
| author = {Cruz, Christopher}, |
| year = {2026}, |
| month = mar, |
| eprint = {2603.27905}, |
| url = {https://arxiv.org/abs/2603.27905} |
| } |
|
|
|
|
|
|
|
|
|
|
| @misc{arxiv260423577, |
| title = {RouteNLP: Closed-Loop {LLM} Routing with Conformal Cascading and Distillation Co-Optimization}, |
| author = {Guo, Dongxin and Wu, Jikun and Yiu, Siu Ming}, |
| year = {2026}, |
| month = apr, |
| eprint = {2604.23577}, |
| url = {https://arxiv.org/abs/2604.23577} |
| } |
|
|
| @misc{arxiv260530898, |
| title = {UniScale: Adaptive Unified Inference Scaling via Online Joint Optimization of Model Routing and Test-Time Scaling}, |
| author = {Huang, Kaiyu and Wang, Xingyu and Kong, Mingze and Shi, Zhubo}, |
| year = {2026}, |
| month = may, |
| eprint = {2605.30898}, |
| url = {https://arxiv.org/abs/2605.30898} |
| } |
|
|
|
|
| @misc{arxiv260312520, |
| title = {When {LLM} Judge Scores Look Good but Best-of-{N} Decisions Fail}, |
| author = {Landesberg, Eddie}, |
| year = {2026}, |
| month = mar, |
| eprint = {2603.12520}, |
| url = {https://arxiv.org/abs/2603.12520} |
| } |
|
|
| @misc{arxiv260417433, |
| title = {Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient {LLM} Reasoning}, |
| author = {Saparkhan, Raman and Hawasly, Majd and Parvez, Md Rizwan and Raza, Mohammad}, |
| year = {2026}, |
| month = apr, |
| eprint = {2604.17433}, |
| url = {https://arxiv.org/abs/2604.17433} |
| } |
|
|
|
|
|
|
| @misc{arxiv260913559, |
| title = {Carbon-Aware Routing for Function Calling in Edge-Cloud {LLM} Systems}, |
| author = {Panteleaki, Aikaterini Maria and Paramanayakam, Varatheepan and Tragoudas, Spyros and Anagnostopoulos, Iraklis}, |
| year = {2026}, |
| month = sep, |
| eprint = {2609.13559}, |
| url = {https://arxiv.org/abs/2609.13559} |
| } |
|
|
| @misc{arxiv260713347, |
| title = {LLM-as-a-Judge Scores Are Unreliable Optimization Signals in Closed-Loop Table Recognition}, |
| author = {Kim, Donghwan}, |
| year = {2026}, |
| month = jul, |
| eprint = {2607.13347}, |
| url = {https://arxiv.org/abs/2607.13347} |
| } |
|
|
| @misc{arxiv260600135, |
| title = {On Effectiveness and Efficiency of Agentic Tool-calling and RL Training}, |
| author = {Liu, Tong and Qian, Cheng and Cief, Matej and He, Yuan and Dan, Daniele}, |
| year = {2026}, |
| month = may, |
| eprint = {2606.00135}, |
| url = {https://arxiv.org/abs/2606.00135} |
| } |
|
|
| % ---------------- External: pre-2026-03 classics (verified 2026-09-23) ---------------- |
|
|
| @misc{arxiv220311171, |
| title = {Self-Consistency Improves Chain of Thought Reasoning in Language Models}, |
| author = {Wang, Xuezhi and Wei, Jason and Schuurmans, Dale and Le, Quoc and Chi, Ed and Narang, Sharanya and Chowdhery, Aakanksha and Zhou, Denny}, |
| year = {2022}, |
| eprint = {2203.11171}, |
| url = {https://arxiv.org/abs/2203.11171} |
| } |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| % ---------------- Benchmark (verified 2026-09-23) ---------------- |
|
|
| @misc{bfcl2024leaderboard, |
| title = {Berkeley Function Calling Leaderboard (BFCL)}, |
| author = {{UC Berkeley (Gorilla project)}}, |
| year = {2024}, |
| url = {https://gorilla.cs.berkeley.edu/leaderboard.html}, |
| note = {Dataset: \url{https://huggingface.co/datasets/gorilla-llm/Berkeley-Function-Calling-Leaderboard}} |
| } |
|
|
|
|
|
|
| @misc{arxiv240417140, |
| title = {Small Language Models Need Strong Verifiers to Self-Correct Reasoning}, |
| author = {Zhang, Yunxiang and Khalifa, Muhammad and Logeswaran, Lajanugen and Kim, Jaekyeom and Bhatt, Gaurav and Zettable, Yulia and Faruqui, Manaal}, |
| year = {2024}, |
| eprint = {2404.17140}, |
| url = {https://arxiv.org/abs/2404.17140} |
| } |
|
|
| @misc{arxiv250618781, |
| title = {Existing {LLMs} Are Not Self-Consistent For Simple Tasks}, |
| author = {Lin, Zhenru and Tao, Jiawen and Yuan, Yang and Yao, Andrew Chi-Chih}, |
| year = {2025}, |
| month = jun, |
| eprint = {2506.18781}, |
| url = {https://arxiv.org/abs/2506.18781} |
| } |
|
|
