GPT-6 Sol leaks, screenshots and release rumors

OpenAI's changelog confirms an image encoding bug had lowered Sol and Luna image understanding across the API, Codex vision tasks and Computer Use; Luna's ReCoG visual grounding rose from 28.8% to 60.0% after the fix, and OpenAI advises rerunning evaluations for affected workflows.

mark_k argues OpenAI is short on GPU capacity right now, so GPT-6 Sol is likely a smaller model replacing Terra to relieve inference load, which explains its lackluster performance.

A Japanese-language morning roundup covers several AI developments, including a fix for an image understanding bug in GPT-6 Sol and Luna and a reset of usage limits.

A poster says testing on the day showed GPT-6 Sol starting to refuse offensive security research requests, including authorized security research.

andonlabs reran Blueprint-Bench for GPT-6 Sol and Luna because of an OpenAI bug: Sol improved slightly while Luna made a large jump, with score charts included.

A long post argues that GPT-6 Sol's half-price launch still didn't bring developers back, since what matters is subscription allowance rather than unit price, and DevDay is unlikely to turn it around.

A comparison post says Sonnet 5.5 performs far better than GPT-6 Sol, and that only Astra 6.1 can save OpenAI from humiliation.

A Japanese post reports OpenAI has fixed a bug that degraded visual tasks in GPT-6 Sol and Luna, improving accuracy via API, Codex, and Computer Use at no extra cost.

A complaint post says OpenAI is clearly prioritizing workflows and Codex, that GPT-6 Sol doesn't feel built for conversation, and that the situation resembles the same time last year.

An OpenAI staffer says an image encoding bug that degraded image understanding in GPT-6 Sol and GPT-6 Luna has been fixed, and invites users to try it that day.

A tipster says GPT-6 Sol is still being developed for Chat mode as fast as possible, that the version in ChatGPT Work, Codex and Text Arena may not be the final checkpoint for conversation, and that current benchmarks shouldn't be treated as final.

Relaying OpenAI's developer account, the image understanding issue in GPT-6 Sol and Luna has been fixed, and Codex computer use was affected too.