OpenAI fixes image-understanding bug affecting GPT-6 Sol and Luna

On September 28, OpenAI's developer account (@OpenAIDevs) announced it had fixed a bug that degraded image understanding in GPT-6 Sol and GPT-6 Luna. After the fix, vision tasks in the API and Codex (including computer use) should improve noticeably; the company advised all workflows using image inputs to re-run evals and try again.

Confirmed

Why it matters

Image understanding is a core capability for multimodal workflows, and this regression directly affected image-dependent automation and agent tasks. OpenAI proactively disclosing the bug and urging users to re-run evals suggests such low-level encoding issues may have significantly dragged down vision performance; users who abandoned or downgraded workflows due to poor image results should reassess after the fix. Positive third-party feedback partly corroborates the fix's effectiveness.

2026-09-28 ~ 2026-09-28 · 6 related posts

Primary sources

2 near-duplicate retellings: OpenAIDevs · OpenAIDevs