Research map / Large language models / Language, safety and interpretability
Is Your LLM Overcharging You? Tokenization, Transparency, and Incentives
Ander Artola Velasco, Stratis Tsirtsis, Nastaran Okati, Manuel Gomez Rodriguez
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase top-p from 0.90 to 0.99 | token-splitting heuristic, Llama-3.2-1B-Instruct · LMSYS T=1.3 | max % tokens overcharged: 0.28 -> 11.2 | Improved |
| raise number of splitting iterations m beyond optimum | token-splitting heuristic · LMSYS T=1.3 | additional revenue | Hurt |
| increase sampling temperature | token-splitting heuristic · LMSYS | fraction of outputs with longer plausible tokenization | Improved |
| raise top-p from 0.90 to 0.99 | misreporting provider · LMSYS T=1.3 | provider utility gain: 0.4 -> 10.5 | Improved |
| lower provider profit margin | misreporting provider · LMSYS T=1.3 | relative utility gain | Improved |
| switch from pay-per-token to pay-per-character with r_c = r_o*tpc | provider profit margin · LMSYS T=1.3 p=0.95 | average profit margin: 92.4%, 93.3%, 94.1% positive margin | No clear effect |
Closest papers by meaning
- Test-Time Compute Games (NeurIPS 2026 · similarity 0.86)
- Fairshare Data Pricing via Data Valuation for Large Language Models (NeurIPS 2025 · similarity 0.85)
- A Theory of Training Profit-Optimal LLMs (NeurIPS 2026 · similarity 0.85)
- Mechanism Design for LLM Fine-tuning with Multiple Reward Models (NeurIPS 2025 · similarity 0.83)
- Pricing and Competition for Generative AI (NeurIPS 2024 · similarity 0.83)
- Incentivizing Quality Text Generation via Statistical Contracts (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
