OpenAI said it has fixed an image encoding bug affecting GPT-6 Sol and GPT-6 Luna, a flaw that reduced the models’ ability to understand images. The company said the fix improves vision-related tasks across the API and Codex, including Computer Use, where an agent reads a screen and operates a computer.
In tests published by OpenAI, GPT-6 Luna (Max) posted a sharp gain on the RefCOCOg visual grounding benchmark. Its average IoU rose from 28.8% to 60.0%, an increase of 31.2 percentage points. The benchmark measures whether a model can identify the correct object or region in an image based on a text description.
Both models only went live in the API on Sept. 22, and OpenAI’s changelog recorded the fix on Sept. 25. The company did not disclose the size of the improvement for Sol. It also did not say when the bug began or how many image requests were affected. OpenAI separately advised developers using image inputs to rerun evaluations and retry workflows that may have been impacted earlier.
OpenAI has fixed an image encoding bug in GPT-6 Sol and GPT-6 Luna, saying the issue reduced the models’ ability to understand images. With the fix in place, vision tasks in the API and Codex should perform better. That includes Computer Use, where an agent reads a screen and operates a computer.
Luna posts a large gain in OpenAI’s test
In results released by OpenAI, GPT-6 Luna (Max) improved its average IoU on the RefCOCOg visual grounding benchmark from 28.8% to 60.0%, a gain of 31.2 percentage points. The test is designed to measure whether a model can accurately locate the object or region in an image that matches a written description.
Fix logged three days after API launch
The two models officially launched in the API on Sept. 22. OpenAI’s changelog then recorded the fix on Sept. 25. The company did not publish a specific improvement figure for Sol, and it did not say when the bug started or how many image requests were affected.
OpenAI also advised developers using image inputs to rerun their evaluations and retry workflows that may have been affected. The company described that recommendation in unusually direct terms.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
