Skip to main content
← All news
ai-models

ChatGPT Puts GPT-5.6 Sol on Instant, Unlimited Luna on Free, and Leaves Work and Codex Alone

By AgentRiot Editorial

OpenAI’s August 6, 2026 ChatGPT update unifies paid Instant and reasoning on a retuned Sol, moves Free and Go to Luna with a later unlimited-text window, and keeps agent surfaces on the July models.

GPT-5.6 Sol Instant ChatGPT update Aug 6 2026
openaichatgptgpt-5.6gpt-5.6-solgpt-5.6-lunacodexchatgpt-work

OpenAI’s August 6 ChatGPT update is not a new frontier-model launch. It is a Chat-stack reset: Plus and Pro get a retuned GPT-5.6 Sol across Instant and deeper reasoning, Free and Go move to GPT-5.6 Luna with a path to unlimited text chats, and the versions of Sol and Luna that power Work and Codex stay on the July release.

The company posted the change as a five-post X thread and backed it with a blog post, ChatGPT release notes, and a 29-page “GPT-5.6 – August Updates” system card dated 2026-08-06.

What changed in Chat

For Plus and Pro, OpenAI says the updated GPT-5.6 Sol now powers both Instant replies and deeper reasoning. The goal is one model personality across speed settings: more focused answers, less unnecessary formatting, detail that scales with the question, and fewer factual mistakes when answers depend on dates, numbers, sources, rules, or assumptions.

Those users also get a reasoning-effort slider on web, mobile, and desktop. OpenAI’s thread presents it as a simpler control than juggling separate Instant and Thinking personalities, with effort labels in the demo running through Instant, Medium, High, and Extra High. The company says paid users can use the updated Sol and slider starting August 6.

For Free and Go, the default everyday model becomes GPT-5.6 Luna. OpenAI’s blog and ChatGPT release notes put that default switch this week. Unlimited text chats and a new Think button for harder questions follow next week, subject to abuse guardrails. File uploads, images, and other tools keep their limits.

That timeline matters because OpenAI’s root X post compressed the free-tier story to “unlimited text chats with GPT-5.6 Luna starting tomorrow.” The blog and release notes are more precise and should be treated as the operational schedule until the product catches up.

Think on free tier is still Luna thinking longer. It is not a free on-ramp to Sol.

The scope boundary that actually matters

OpenAI is explicit about where this does not land:

  • The updated Sol is for everyday Chat only.
  • ChatGPT Work and Codex keep the previously released (July) GPT-5.6 Sol and Luna.
  • The August system card labels the new Chat models as the August Sol/Luna pair and the unchanged agent/coding pair as July.

If you build agents, plugins, or coding workflows, do not assume today’s Chat retune changed your Work or Codex model behavior, pricing, or eval baseline. It did not, per OpenAI.

Also note a documentation lag: as of August 6 access, OpenAI’s older Help Center article “GPT-5.6 in ChatGPT” still described Instant as GPT-5.5 Instant and still framed Free/Go without GPT-5.6 Sol access. That page had not been rewritten to match the new Instant-on-Sol design. Prefer the August 6 blog, release notes, and system card over that stale picker table until Help Center catches up.

Vendor factuality claims, with the labels they need

OpenAI’s headline number, repeated on X, is that in an internal high-stakes factuality evaluation covering finance, medicine, and law, the new GPT-5.6 Sol produced 68% fewer responses with at least one factual error than GPT-5.5 Instant. The blog adds that GPT-5.6 Luna cut those error-containing responses by about 62% on the same class of prompts.

The August system card widens the picture and also narrows how far the numbers should travel:

  • The hallucination sets are deliberately hard (factuality-heavy production-like prompts, user-flagged prior failures, and high-stakes medical/legal/financial prompts).
  • They are not estimates of average production error rates.
  • Graders are LLM-based with web access.
  • OpenAI reports Sol cutting factual error rates by roughly 60% across the three sets, with Luna above 60% on high-stakes and about 30% on the other two.

Treat all of that as OpenAI-reported, not independently reproduced. Useful directional signal. Not a third-party leaderboard result.

On HealthBench, the system card’s length-adjusted scores put August Sol above GPT-5.5 Instant on the main HealthBench suite and especially on HealthBench Professional (54.0 vs 38.4 length-adjusted). Same caveat: vendor eval, length-adjusted methodology disclosed in the card.

Safety package, not a blank check for unlimited chat

OpenAI says the August Sol and Luna Chat models keep the same Preparedness Framework posture as the prior GPT-5.6 family: High in Biological and Chemical, High in Cybersecurity, and below High in AI Self-Improvement, with the same safeguard package described for GPT-5.6.

The blog and system card also highlight first-time dedicated under-18 evaluations and additional teen-oriented training and system protections: avoiding romantic roleplay, age-restricted challenges, and positioning the model as a substitute for real-world relationships; tighter boundaries on sexual content, eating disorders and body-image risk, age-restricted goods, dangerous activities, and graphic violence; and nudges toward trusted people when a teen may need support. Those are OpenAI’s stated design goals and internal scores, not external audits of real-world outcomes.

Unlimited free text chat widens exposure even with those guardrails. OpenAI itself still ties free unlimited text to abuse controls and keeps non-text tools limited.

How to read the product move

Three concrete shifts:

  1. Paid Chat collapses Instant and Thinking onto one Sol. The control surface becomes effort, not model family switching, for ordinary Chat.
  2. Free Chat gets more volume on Luna, not Sol. Unlimited text is a capacity and retention move. The Think button is extra Luna compute, not free flagship access.
  3. Agent surfaces are intentionally frozen. Work and Codex stay on July Sol/Luna while Chat gets the August retune. That split is the cleanest operator takeaway.

OpenAI frames the release around “more abundant intelligence” and says a billion people use ChatGPT every week. The adoption figure is company-reported. The more durable product fact is narrower: ChatGPT’s default consumer path is being re-tiered around Sol for paid Instant+reasoning and Luna for free scale, while the serious work surfaces keep yesterday’s model pins.

What to watch next

  • Whether Help Center model-picker docs update to match Instant-on-Sol for Plus/Pro.
  • When Free/Go actually receive Luna default, unlimited text, and Think, relative to the “this week / next week” schedule.
  • Whether July Work/Codex Sol later receives the same factuality/focus retune, or stays a separate train.
  • Independent checks of the high-stakes factuality claim; none are public with this launch package.

Until those land, the accurate story is an August 6 ChatGPT Chat update with a clear non-scope: better paid Sol in Chat, broader free Luna access on a short delay, and no simultaneous Work or Codex model change.

Sources