ChatGPT free tier goes unlimited. Reprice your AI features.
OpenAI removed text rate limits for free ChatGPT users and shipped a reasoning-effort slider for paid ones. Two signals for how you price and build AI.
On August 6, OpenAI removed text rate limits from the ChatGPT free tier entirely and made GPT-5.6 Luna the default model for Free and Go users. On the paid side it shipped a slider that lets people set how hard the model thinks. Both moves are pricing signals, and both should change how you build AI features into your own product.
What actually happened
Per TechCrunch and MacRumors:
- Free and Go users get unlimited text chats — no text-based rate limits, subject to abuse guardrails.
- GPT-5.6 Luna replaces GPT-5.5 Instant as the Free and Go default.
- Limits remain on file uploads, images, voice, and image generation. Text is the thing that went unmetered.
- A Think button gives free users deeper reasoning on a per-message basis, rolling out the week of August 10.
- Plus and Pro get an updated GPT-5.6 Sol plus a thinking slider on web, mobile, and desktop — quick replies at one end, extended effort for planning and coding at the other.
- OpenAI reports factual errors down 62% for Luna and 68% for Sol against GPT-5.5 Instant.
Why AI pricing matters for your business
Unmetered text means the marginal cost of a cheap-model round trip is now rounding error. OpenAI does not give away a billion-user product at a loss it can't model. If a lightweight text call is cheap enough for OpenAI to stop counting, it is cheap enough for you to stop gating. If you built a credit system around basic text generation in 2025, that pricing page is now a competitive liability. Meter the expensive things — vision, long context, deep reasoning, document parsing — and stop nickel-and-diming a completion that costs you a fraction of a cent.
The slider is the pattern worth copying. OpenAI just made reasoning effort a user-facing control instead of a hidden config value. That is the right shape for most AI features: default to fast and cheap, expose an explicit "work harder on this one" affordance, and let the user spend your compute deliberately. It caps your cost curve and it sets expectations about latency without a spinner apologizing for eleven seconds.
Two model tiers, one product. Luna for the default path, Sol for the escalation path. Route by task, not by customer plan. Build the routing layer yourself so the model names are configuration, not architecture — because Luna and Sol will not be the names in six months.
Key takeaways
- ChatGPT Free and Go now have unlimited text chats; files, images, and voice stay capped
- GPT-5.6 Luna is the new free default; Think button arrives the week of August 10
- Plus and Pro get updated GPT-5.6 Sol plus a user-facing reasoning-effort slider
- OpenAI reports 62% (Luna) and 68% (Sol) fewer factual errors vs GPT-5.5 Instant
- Stop metering cheap text in your own product; meter vision, long context, and deep reasoning
- Expose an effort dial instead of picking one speed for every user
Cheap-tier default, expensive-tier escalation, your routing logic in the middle. We build AI features with the model layer behind a config file, so a price cut or a model retirement is a one-line change instead of a rebuild. See how we architect it, or run the numbers on what your AI features actually cost per task.
Sources: TechCrunch, MacRumors.
- #openai
- #chatgpt
- #ai-pricing
- #llm-costs
- #product-design
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Suno will watermark AI songs and cap your downloads
Suno is adding watermarking, fingerprinting, and download limits to AI-generated music. Your rights to distribute AI output are a vendor setting, not a deed.
Read itSapiom's $35M: your agent bill is a routing decision
Sapiom raised $35M Series A to route AI agent calls to the cheapest capable model. The lesson isn't the vendor — it's that model choice belongs in config, not code.
Read it