Codex Radar is an information and tools site that tracks Codex and ChatGPT subscription quotas and model quality status. Its core purpose is to help users keep up with quota reset progress, judge whether a model has experienced quality degradation, and compare the performance of different model tiers in benchmark tests.
In terms of quota monitoring, the site records every direct reset and reset-card issuance in Beijing time, distinguishes between the two event types of direct reset and card issuance, and marks the differences in status among official previews, official confirmations, and site-owner confirmations. For events whose per-account crediting or official completion post has not yet been verified, the site clearly states that the actual landing time is unknown. The site also provides a reset time distribution chart, summarizing the hourly distribution of historical reset/card-issuance events, but notes that historical patterns do not guarantee future reset times.
In terms of model quality monitoring, the site uses the DeepSWE benchmark and cached interfaces to score multiple model tiers, covering more than 40 different questions, sorts them by radar score, and displays reference data such as IQ, time consumption, and average API-equivalent cost. The site also lists two community open-source projects for model fingerprint detection and Codex model change monitoring, and notes that the results are for reference only and cannot be used to determine the actual model being used based on the results alone.
In terms of practical experience, the site compiles multiple usage suggestions, including using local MCP services with the official tunnel on the web version of ChatGPT Pro to collaborate with local Codex, context setting suggestions for Sol and Astra, the impact of switching reasoning intensity on caching and cost, applicable scenarios for Ultrafast, how to enable the input prediction feature, and reset cards and quota conversion, etc. Some content is marked as tested by the site owner, based on community discussion standards, or personal feedback, and is clearly stated not to be controlled testing.
The site provides community prediction and scoring features. Users can rate model tiers and participate in daily predictions, but JavaScript must be enabled. Interactive content has nothing to do with real quotas. Overall, Codex Radar provides users who follow Codex quota cycles and model status with a centralized entry point for viewing official updates, community feedback, and reference data.
