OpenAI is reportedly testing an “Aurora” image model on Arena, while a paper finds only partial agreement between API and ChatGPT citations.
OpenAI tracker4 items
A Python SDK patch fixes Realtime compatibility, and a new benchmark highlights limits in agents’ strategic decisions.
OpenAI reportedly tests “Aurora” GPT-Image model on Arena
TestingCatalog reports that OpenAI is testing a new GPT-Image model called “Aurora” on Arena. The item provides no release timing or availability details.
Why it matters: This signals potential movement in OpenAI’s image lineup, but does not establish a product launch.
Paper finds OpenAI API citations only partly match ChatGPT’s
A paper covered by Search Engine Journal finds that OpenAI’s API only partly matched ChatGPT’s citations. The item does not specify the models, methods or size of the discrepancy.
Why it matters: Developers evaluating citation behavior should not assume API results mirror the ChatGPT experience.
Python SDK 3.35.1 fixes beta Realtime byte reads
OpenAI’s Python SDK version 3.35.1 fixes a bug to preserve beta Realtime byte reads when using websockets 13.
Why it matters: The patch addresses compatibility for developers using the affected Realtime and WebSocket configuration.
GPT-5.6 Sol scores below 60% on Taste-Bench strategic choices
Taste-Bench tests AI agents’ strategic choices during long-running tasks. GPT-5.6 Sol, the best-performing model reported, chose correctly on just under 60% of simple binary questions; longer thinking did not help.
Why it matters: The results flag strategic decision-making as a limitation when evaluating agents for long-running workflows.