OpenAI Codex investigates billing and usage anomalies

// 2h agoNEWS

OpenAI Codex investigates billing and usage anomalies

Thibault Sottiaux of the OpenAI Codex team announced that engineers are working over the weekend to investigate reports of abnormal usage drains affecting some users. In addition, Sottiaux refuted claims that the platform routes prompts to cheaper models, clarifying that the Codex Auto-review system runs directly on GPT-5.4 without any routing mechanisms.

// ANALYSIS

Billing and token-metering discrepancies are critical concerns for developers running agentic workflows, making swift, transparent communication necessary to maintain user trust.

* Rapidly investigating "usage drains" prevents developers from incurring massive, unexpected costs when executing automated loops.

* By clarifying that Auto-review is using GPT-5.4 and stating "We do not route," the team attempts to counter suspicions of bait-and-switch downgrades to cheaper models.

* Weekend "warroom" responses emphasize how highly OpenAI prioritizes reliability and fair billing for its enterprise-oriented developer tools.

// TAGS

openaicodexgpt-5.4code-reviewrate-limitingdebuggingagentai-codingcode-generation

DISCOVERED

2h ago

2026-06-28

PUBLISHED

2h ago

2026-06-28

RELEVANCE

7/ 10

AUTHOR

thsottiaux

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

INFRA41m ago

CuPy provides GPU-accelerated drop-in replacement for NumPy and SciPy

CuPy is an open-source array library for GPU-accelerated computing with Python, providing highly compatible API replacements for NumPy and SciPy. By leveraging CUDA Toolkit libraries and AMD ROCm, CuPy allows data scientists and researchers to execute complex numerical algorithms and data manipulation tasks significantly faster than CPU implementations, without requiring extensive rewrites of their existing Python code. It has become a staple in the scientific computing ecosystem, currently trending on GitHub with over 11,500 stars.

RESEARCH1h ago

Researchers introduce DiffusionBench for holistic DiT evaluation

Researchers argue that current Diffusion Transformer (DiT) evaluations over-rely on ImageNet, which poorly correlates with real-world text-to-image performance. To address this, they introduce NanoGen for unified training and DiffusionBench, a holistic benchmark for evaluating DiTs across both tasks.

NEWS1h ago

GLM-5.2 launch accelerates green energy transition

Alan Kohler's latest column connects geopolitical events in the Strait of Hormuz with the release of Zhipu AI's GLM-5.2 model. While Iranian maritime pressure drives up fossil fuel costs, the power demands of open-weight Chinese AI models accelerate the global transition to renewable energy.