AI Models
27 September 2026
OpenAI fixes image recognition error in GPT-6
N
New-ZZZ desk
X @OpenAIDevs · 1 day ago
OpenAI has eliminated a critical bug that previously degraded the ability of GPT-6 Sol and GPT-6 Luna models to understand images. Users should now receive significantly better results when performing visual tasks via the API and Codex.
This improvement is especially noticeable in tasks related to recognizing computer usage and other complex visual scenarios.
Developers are strongly recommended to rerun their tests and workflows (evals) to ensure stability and quality after the fixes.
Why it matters
- —Improved visual understanding is critical for complex workflows requiring image analysis.
- —The fix increases the reliability of API usage, making it more stable for computer vision tasks.
- —This makes GPT-6 models more versatile and accurate when dealing with the real world.
Key facts
- An error that degraded image understanding has been eliminated.
- GPT-6 Sol and GPT-6 Luna models have been fixed.
- Improvements are available for use via API and Codex.
- Developers are recommended to rerun tests (evals).
The full text is in the original source. Here we provide a brief summary and key facts.
