ChatGPT Splits Free Unlimited Luna From Paid Sol — Lab Factuality Only
OpenAI retuned GPT-5.6 Sol for ChatGPT Plus/Pro with a thinking-effort slider and made Luna the Free/Go default with unlimited text chats. Internal evals claim ~68% / ~62% fewer factual-error responses vs GPT-5.5 Instant — company-run, not Arena or Artificial Analysis. Distribution day, not a new base-model family.
OpenAI is splitting the free tier into unlimited cheap intelligence (Luna text) while selling deeper, cleaner Sol behind Plus/Pro — with a slider that productizes “how hard should the model think?” as UX instead of model-picker jargon.
OpenAI updated ChatGPT defaults across paid and free tiers. Plus and Pro get a ChatGPT-tuned GPT-5.6 Sol (separate from the Sol used in Codex / ChatGPT Work) plus a thinking-effort slider that replaces separate Instant vs Thinking modes. Free and Go users get GPT-5.6 Luna as default, with unlimited text chats and a Think button rolling out on a staggered schedule. OpenAI says one billion people now use ChatGPT weekly.
What changed by tier
| Tier | Model default | New controls | Notes (OpenAI + release notes) |
|---|---|---|---|
| Plus / Pro | Updated GPT-5.6 Sol (ChatGPT path) | Thinking slider | More focused answers; less excess formatting; adapts detail to question difficulty |
| Free / Go | GPT-5.6 Luna | Think button for harder questions (rolling) | Unlimited text chats next (abuse guardrails remain); limits stay on files, images, voice, image gen, other tools |
| Codex / Work Sol | Unchanged | — | ChatGPT Sol retune is product-path specific |
Claims vs checks
OpenAI reports an internal evaluation on finance, medicine, and law prompts requiring factual detail:
| Model | Claim vs GPT-5.5 Instant |
|---|---|
| GPT-5.6 Sol (updated ChatGPT) | ~68% fewer responses with ≥1 factual error |
| GPT-5.6 Luna | ~62% fewer responses with ≥1 factual error |
These are company-run figures — not independent arena or third-party factuality boards. Treat as lab-reported until Artificial Analysis / Arena-style audits catch up.
Not a new base-model family GA — GPT-5.6 Sol/Terra/Luna already public from the July family launch and subsequent price cuts. Independent LMArena / Artificial Analysis ranks for the original Sol/Luna tiers still apply as the backbone; this post does not publish a new public Elo. Note when ChatGPT-path Sol diverges from API/Codex Sol in user experience.
Limits
- Factuality cuts are internal only; no new public Elo.
- Unlimited free text may raise abuse/cost pressure — not measured here.
- Work/Codex Sol stays deliberately different from chat Sol.
Sources
- OpenAI: Improving GPT-5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free users (August 6, 2026)
- OpenAI ChatGPT release notes (August 6, 2026 section)
- The Decoder: OpenAI improves GPT-5.6 Sol and free-tier Luna shift (August 6, 2026)
- Times of AI:
openai-gpt-5-6-luna-terra-price-cuts(July 30)