Token Pollution (V3) – Addendum Cross‑validation against OpenAI's official pricing (GPT‑5.4, Sora‑2) – 22% error margin, zero internal data
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This document is an addendum to the Token Pollution V3 report. It compares the estimates of video generation costs (Sora) with the official pricing grid published by OpenAI on 27 March 2026 (GPT‑5.4, Sora‑2, GPT‑image‑1.5, GPT‑realtime‑1.5). The comparison shows a 22 % margin of error between the external estimate (€1.12 for 10 seconds of video) and the official price (€0.92), confirming the robustness of the methodology. This document is available in both English and French. Methodology across V2, V3, and this addendum All three reports were produced without any internal data from OpenAI, using only publicly available sources: academic energy ratios (Carnegie Mellon, CodeCarbon), iOS download figures, and press reports. The work was carried out entirely on a mobile phone, with no dedicated computing infrastructure. The process relied on iterative orchestration of AI models (DeepSeek, Claude, Grok) for quantitative modelling, cross‑verification, and document structuring. No automated scripts or spreadsheets were used, all calculations and scenario matrices were performed through mental modelling and manual cross‑checks. Timeline and publication • Token Pollution V2 (March 7, 2026) established the energy ratios and the concept of contextual pollution. • Token Pollution V3 was released two days after OpenAI announced the closure of Sora (24 March 2026). • This addendum was written the day after OpenAI published its official pricing (27 March 2026), validating the V3 estimates with a 22 % margin. Contents • Full OpenAI pricing grid (March 2026) • Detailed cost calculations per second and per video format • Factor ×8 conversion from electricity cost to operator cost • Cross‑validation tables and scenario comparisons with V2/V3 Related reports • Token Pollution V2: https://zenodo.org/records/18904684 • Token Pollution V3: https://zenodo.org/records/19242735
Fields of Study & Subjects
Green IT and Sustainability
Artificial Intelligence Applications
Energy Efficiency in Computing
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 28 March 2026
- Language
- EN