GPT-5.6 Sol Launch, Fable Resets, Codex Subagents — AI Daily Jul 09
898 messages · 83 active members
Overview
Topics
OpenAI shipped 5.6 in three tiers over a 24-hour window with no auto-update — members had to force `codex update` or npm reinstall. Sol delivers 372K oAuth context (1.5M API), plans deeper and asks more clarifying questions than 5.5/Fable, but is slower. Luna isn't actually fast yet (waferscale pending). The 'ChatGPT Work' livestream landed as a Codex rebrand plus a genuinely improved GPT-5.5 voice mode.
Anthropic pushed a surprise mass reset of Fable usage mid-day, zeroing out members at 90%+ utilization — widely read as defensive counter-programming during the 5.6 launch. Fable still catches issues 5.5 misses on complex CRM refactors, but multiple builders confirmed the post-launch nerf, with Opus 5 rumored as the next drop.
Codex shipped a new subagents UI with per-agent visible windows and a browser that imports Chrome profiles/cookies (Brave not supported). Sol Ultra/Fast modes are burning limits 2x faster than 5.5 — one prompt reportedly nuked a Pro plan in a single turn. Builders are switching to terminal Codex over the desktop app and dropping to High effort to survive.
Grok 4.5 landed on the Artificial Analysis intelligence frontier, though @jcartu flagged confident bullshitting as a harder failure mode than hallucination. Consensus playbook: Sol for planning/reviews, Terra for cheaper execution, Codex for fast coding (~90% of tokens), Opus/Fable for docs and UI. @ltv_1 detailed a Hermes MoA setup where a main GPT-5.5 profile dispatches to specialist sub-profiles.
Members shared one-shot video workflows with Grok Imagine, Seedance (praised for ambient audio), LTX 2.3, and Wan 2.2 — with room-tone prompting as the current frontier. @RayTGcontact shared a CPL media buying bot suggesting bid caps and pushing winners to a creative bot, and @justingacina's framework: start with human y/n approval, auto-approve the 99% patterns. @fmill1 detailed acquiring a $3K MRR product for $500 via FounderBase.
Key Takeaways
- GPT-5.6 requires manual `codex update` or npm reinstall — no auto-prompt. Sol is the reasoning default; context jumps to 372K oAuth / 1.5M API.
- Sol Ultra/Fast burns tokens 2x faster than 5.5 — one prompt can nuke a Pro plan; drop to High effort and switch to terminal Codex over the desktop app.
- Anthropic mass-reset Fable usage mid-launch as defensive counter-programming; Opus 5 rumored next. Fable still wins complex refactors despite post-launch nerf.
- Grok 4.5 hit the intelligence frontier but bullshits more confidently — a harder failure mode than hallucination for 'truth-telling' model claims.
- Emerging playbook: Sol for planning, Terra for cheap execution, Codex terminal for coding (~90% of tokens), Fable/Opus for UI and docs. Luna isn't fast yet.
Hot Threads
GPT-5.6 rollout mechanics, Sol/Terra/Luna testing, and ChatGPT Work reactions
Fable usage reset discovery, nerf theories, and Opus 5 speculation
Media buying bot: bid caps, approval loops, and creative handoff