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.

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.

The author fed a muted video to Codex (GPT-6 Astra/Sol) to design sound effects without a DAW or third-party samples, and shared a reusable prompt.

The author shows Codex (GPT-6 Sol) using tmux capture-pane to peek at Opus 5.5's progress and tmux send-keys to talk to it, and taking over unfinished work when Claude runs out of tokens.

A leaker speculates astra-minor is a distilled Astra that could fill the pricing gap between GPT-6 Sol ($2/$10) and Astra ($10/$50), competing with Opus 5.5 at $4/$20.

A user who bought Codex Plus to compare GPT-6 Sol with Astra says the 5-hour limit ran out in 10–15 minutes without finishing one prompt, and threatens to cancel.

A developer used GPT-6 Sol, Codex and Three.js to turn a 2020 craft cluster trip to Jaipur into an interactive fictional neighborhood, with a live demo online.

The update switches Devin from devin acp to Devin's own API, reusing CLI login with no extra setup, and has been tested with Claude Code and OpenCode on Opus 5.5, SWE-2 and GPT-6 Sol for multi-turn file reads and writes.

The poster notes 5.6 Terra was removed from the official usage limits page after GPT-6 Sol launched, and that Sol's 5-hour limit of 15-150 is lower than Terra's 25-200 despite identical API pricing.

A user asks whether Codex is down, says Claude is doing well with Opus 5.5, and notes OpenAI is juggling model and capacity issues.

A post criticizing OpenAI says GPT-6 Sol feels weaker than 5.6 Sol on many real coding tasks and Luna is noticeably dumb, while also complaining that quotas burn faster and resets are scarcer.

The demo uses WebMCP to sync video playback state so GPT-6 Sol comments live as the video plays, rather than summarizing subtitles afterward.