Verdict: OpenAI made GPT-5.6 Luna the default model for free ChatGPT accounts on August 6, 2026, with unlimited text chats and no daily message cap for text-only conversations. Free users can now access a frontier-family model at zero cost — but only for text. If you need deep reasoning (GPT-5.6 Sol), image generation, file uploads, or agentic coding in Codex, those still require a paid plan ($20/month Plus or higher). The most cost-effective strategy for most people is to use the free Luna tier for everyday work and pay $20/month only when you need Sol-grade reasoning to plan and orchestrate an hard problem.
Last verified: 2026-08-07
- Free ChatGPT users get unlimited text chats on GPT-5.6 Luna (no daily cap on text) as of August 2026.
- GPT-5.6 Luna is a fast and affordable model in the current frontier family, costing $0.20/1M input and $1.20/1M output tokens on the API — down 80% from its launch price. (OpenAI, OpenAI Pricing)
- GPT-5.6 Sol (the flagship reasoning model) drives the medium/high/extra-high thinking modes for Plus ($20/mo) and Pro ($200/mo) users. Free users get a "Think button" tap for harder questions.
- This is a volatile facts article — pricing, limits, and model naming change frequently. Re-check before making decisions.
What Did OpenAI Actually Change for Free Users on August 6, 2026?
OpenAI's August 6 announcement did two things simultaneously: it improved the paid reasoning experience and expanded what free users can do. For free and Go-tier users, the default model was switched from GPT-5.5 Instant to GPT-5.6 Luna, and text chats were made effectively unlimited (subject to abuse guardrails). A new "Think button" lets free users trigger higher reasoning on harder questions. (OpenAI Blog)
Crucially, the unlimited access is text-only. File uploads, image generation, web browsing, and other non-text tools retain their previous rate limits. If you mostly use ChatGPT for writing, brainstorming, answering questions, drafting emails, and coding explanations — you are now unbounded. If you rely on vision, DALL-E, or file analysis, the free tier still throttles those.
For Plus and Pro users, the same update unified the Chat experience around an improved GPT-5.6 Sol: answers are more direct, factual errors are significantly reduced, and a new reasoning-depth "slider" lets you choose between quick answers and deep planning. The slider replaces the old Instant/Thinking split with a continuous control. (OpenAI Blog)
How Good Is GPT-5.6 Luna Compared to the Paid Sol Model?
GPT-5.6 Luna is OpenAI's fastest and cheapest tier in the current frontier family — roughly corresponding to the "nano" tier in earlier generations. It is not the flagship reasoning model; that would be GPT-5.6 Sol. But Luna is a far cry from a "dumb" model, because it shares the GPT-5.6 architecture and training lineage. (OpenAI API Docs, OpenAI GPT-5.6 Launch)
Here is what the data shows:
| Metric | GPT-5.6 Luna | GPT-5.6 Terra | GPT-5.6 Sol |
|---|---|---|---|
| API input price (per 1M tokens) | $0.20 | $2.00 | $5.00 |
| API output price (per 1M tokens) | $1.20 | $12.00 | $30.00 |
| Role | Fast, cheap, high-volume | Balanced everyday work | Flagship reasoning + agentic coding |
| ChatGPT plan availability | Free (as default) + all paid | Paid (Work/Codex) | Plus and higher (medium/high reasoning) |
| Factual error reduction vs GPT-5.5 Instant | ~62% fewer errors | — | ~68% fewer errors |
| Context window | 1.05M tokens | 1.05M tokens | 1.05M tokens + Sol Ultra mode |
Sources: OpenAI API Docs, ProPakistani, OpenAI Blog (Aug 6)
Key takeaway: Luna's internal evaluation showed 62% fewer factual errors than the previous GPT-5.5 Instant model it replaces as default, per OpenAI's own testing on financial, medical, and legal prompts. So while Sol is the deeper reasoner, Luna is a meaningful upgrade over what free users had before.
The model also scales in intelligence with effort level. Luna at "max" effort is substantially smarter than Luna at "low" effort — from OpenAI's coding index, the gap between low/medium/high/max on Luna is large. Independent benchmarks indicate that DeepSeek V4 Flash, another very cheap model, is roughly equivalent to Luna on max effort. (OpenAI GPT-5.6 benchmarks)
How Can a Free User Get GPT-5.6 Sol Reasoning Without Paying?
Free users get a "Think button" in the ChatGPT interface. Tapping it promotes the query to a higher-reasoning pass. OpenAI has not publicly disclosed whether the free Think button routes to GPT-5.6 Sol specifically, or to Luna at higher effort, but the feature's purpose is to handle harder questions without requiring a subscription. (OpenAI Blog, Android Headlines)
The reality check: if you need sustained, deep reasoning for complex coding plans, multi-step agentic work, or long research sessions, the free Think button provides a tap — not unlimited access to Sol. For consistent high-effort reasoning, the $20/month Plus plan is the entry point.
What Does GPT-5.6 Cost on the API, and Why Was Luna's Price Cut 80%?
On July 30, 2026 — just three weeks after GPT-5.6 went fully public — OpenAI cut Luna's API price by 80% and Terra's by 20%, while Sol stayed unchanged. Luna went from $1/$6 per 1M input/output to $0.20/$1.20. Terra went from $2.50/$15 to $2/$12. (ProPakistani, FoneArena, OpenAI Blog)
OpenAI attributed the cuts to improvements in their inference systems, hardware utilization, production software, and context management. GPT-5.6 Sol itself reportedly helped optimize production systems, reducing serving costs by 20% and improving token-generation efficiency by over 15%. (OpenAI Blog)
This price cut is what made free unlimited Luna chats economically possible. Serving a billion weekly active users on a frontier model for free requires the model to be extraordinarily cheap to run — and at $0.20/1M input tokens, Luna is now among the cheapest frontier-family models on the market.
How Does Free GPT-5.6 Luna Compare to Other Cheap or Free AI Models?
If you are choosing between free / cheap AI options in August 2026, here is a practical comparison:
| Model | Provider | Price (per 1M tokens, in/out) | Open Weights? | Context | Best Free Option? |
|---|---|---|---|---|---|
| GPT-5.6 Luna | OpenAI | $0.20 / $1.20 | No (proprietary) | 1.05M | Yes — unlimited text in ChatGPT free tier |
| DeepSeek V4 Flash | DeepSeek | $0.14 / $0.28 | Yes (MIT) | 1M | Free via DeepSeek app; self-hostable |
| GPT-5.6 Terra | OpenAI | $2.00 / $12.00 | No | 1.05M | No (paid plans only) |
| Claude Opus 5 | Anthropic | $5.00 / $25.00 | No | 1M | No (Pro/Max only) |
| Muse Spark 1.2 | Meta | $1.25 / $4.25 | No (contributor tier $0.10/$0.20) | 1.05M | No (API only) |
Sources: OpenAI, DeepSeek/HuggingFace, Anthropic, CloudPrice
For a free, no-setup experience, ChatGPT with unlimited Luna text is now the most accessible option. For developers who want open-weight deployment or the lowest token price possible, DeepSeek V4 Flash ($0.14/1M input, MIT-licensed) remains the cost leader you can self-host.
For a deeper model-by-model comparison of the current frontier options, see our Qwen3.8-Max vs Claude Fable 5 comparison which breaks down the capabilities and cost of China's latest flagship against Anthropic's top tier.
How to Build a Cost-Efficient AI Workflow Around Free GPT-5.6 Luna
The smartest way to use AI in August 2026 is not to pick one model — it is to tier your work. Here is a practical framework that works for developers, small business owners, and builders:
Step 1: Use free ChatGPT Luna as your everyday workhorse
For brainstorming, writing first drafts, answering research questions, drafting emails, and explaining concepts, the free ChatGPT tier with unlimited Luna text is sufficient. Set it as your default and stop thinking about message counts.
Step 2: Escalate to Sol ($20/month Plus) for planning and orchestration
When you have a genuinely hard problem — a complex coding architecture, a multi-step business strategy, a research synthesis — use GPT-5.6 Sol. The $20/month Plus plan gives you access to medium/high reasoning modes for a monthly budget. A practical pattern: ask Sol to create a plan, then hand the execution work back to Luna (or a cheaper model like DeepSeek V4 Flash) to actually do it.
This "plan with Sol, execute with Luna" approach means most of your token spend goes to cheap Luna calls, with Sol reserved for the infrequent moments where its reasoning genuinely matters.
Step 3: Use open-weight models for bulk or sensitive work
If you are processing large volumes of text (classification, extraction, summarization of documents), or if data privacy is a concern and you cannot use a US-hosted API, consider self-hosting DeepSeek V4 Flash. At $0.14/1M input tokens via API, or free if you run the open weights yourself, it is cheaper than Luna and has no data-retention concerns. The trade-off: you need GPU infrastructure or a serving provider.
For those concerned about enterprise data security when using any AI provider, our analysis of AI models breaching companies during cybersecurity tests covers what the frontier labs' own red teams found.
Step 4: Reserve Sol Ultra / Claude Opus 5 for the hardest problems only
GPT-5.6 Sol Ultra ($30/1M output on the API, available to Pro/Enterprise in ChatGPT) and Claude Opus 5 ($25/1M output) are the top reasoning models. Use them only when the task complexity justifies the 25-30x price premium over Luna. For most everyday work, this is wasteful.
What This Means for You
If you are a free ChatGPT user: You just got a significant upgrade. GPT-5.6 Luna is more capable and more accurate than the model it replaced, and you no longer hit a daily text-chat cap. Use it freely for writing, research, brainstorming, and learning. The main remaining limits are on image generation, file uploads, and vision — if you need those regularly, Plus is worth it.
If you are a small business owner: The free tier is now viable for real work — drafting marketing copy, answering customer FAQs, generating intake form templates, and research. For more complex tasks like analyzing financial documents or generating business plans, one $20/month Plus account on your team gives access to Sol reasoning. You probably do not need more than that unless you are running coders on it all day.
If you are a developer: The Luna price cut to $0.20/1M input tokens makes it the cheapest proprietary frontier-family model for routing, classification, and high-volume API calls. Combined with Sol for orchestration and an open-weight model like DeepSeek V4 Flash for bulk, you can build a three-tier model stack that costs cents per day for personal projects. For agentic coding specifically, our guide on building with agent operating systems covers how to orchestrate multiple models with shared memory.
If you are worried about the economics: OpenAI is reportedly burning billions serving a billion weekly users at a negative operating margin (Memeburn analysis, TechCrunch). Free unlimited Luna is a land-grab play, not a charity — OpenAI is betting that scale, efficiency improvements, and eventual conversion to paid plans will cover the cost. For users, this means free access now is a good deal that may not last forever. Take advantage of it while it stands. For a broader look at whether the AI industry's spending binge will pay off, see our analysis of big tech's AI capex depreciation risk.
FAQ
Q: Is ChatGPT's GPT-5.6 Luna really free with no limits?
A: Free ChatGPT users get unlimited text chats on GPT-5.6 Luna as of August 2026. The limit is on non-text features — file uploads, image generation, web browsing, and vision retain their previous rate limits. Abuse guardrails may still throttle extreme usage. (OpenAI Blog)
Q: Can free users access GPT-5.6 Sol for reasoning?
A: Free users get a "Think button" that promotes harder questions to higher-reasoning mode. Whether this routes to Sol specifically or to Luna at higher effort is not publicly documented. Consistent access to Sol's medium/high/extra-high reasoning modes requires a Plus ($20/mo) or higher subscription. (OpenAI Blog)
Q: How much does GPT-5.6 Luna cost on the API?
A: After the July 30, 2026 price cut, GPT-5.6 Luna costs $0.20 per million input tokens and $1.20 per million output tokens — an 80% reduction from its launch price of $1/$6 per 1M. (ProPakistani, OpenAI)
Q: Is GPT-5.6 Luna better than DeepSeek V4 Flash?
A: They serve different needs. Luna is proprietary and accessible free via ChatGPT; DeepSeek V4 Flash is open-weight (MIT license), cheaper on the API ($0.14/$0.28 per 1M), and self-hostable. On benchmarks, Luna appears roughly comparable to V4 Flash at effort "max" for general tasks. For production cost optimization, V4 Flash wins; for convenience and zero-setup, Luna wins. (OpenAI, HuggingFace)
Q: What is the difference between GPT-5.6 Luna, Terra, and Sol?
A: Luna is the fastest/cheapest tier ($0.20/1.20 per 1M), positioned for high-volume everyday work. Terra is the balanced middle ($2/12), close to GPT-5.5 quality at 60% less cost. Sol is the flagship ($5/30) for agentic coding and the hardest reasoning, with a Sol Ultra mode for maximum effort. All three share a 1.05M-token context window. (OpenAI, OpenAI API Docs)
Q: Should I pay for ChatGPT Plus if I already get free unlimited Luna?
A: Only if you regularly need one of these: (1) GPT-5.6 Sol's deep reasoning for coding, planning, or research, (2) image generation, (3) file upload/analysis, (4) knowledge/vision tools beyond a text chat. If your use is mostly text-based Q&A and drafting, the free tier now covers that without limits. (OpenAI Blog)
Q: What happens to my data when I use free ChatGPT?
A: OpenAI's policy for consumer ChatGPT historically allows the use of conversation data to improve models, with an opt-out setting. For API calls, OpenAI does not use customer data for training by default. If zero data retention is a requirement, consider open-weight models like DeepSeek V4 Flash that you self-host. We cover AI model security and data handling concerns in our AI model cybersecurity testing analysis.
Every claim here is traced to a primary source, dated, and listed under Sources. Research and drafting are AI-assisted; editing, verification and publication are human decisions, and a person is accountable for what appears on this page. How we work →
Discussion
0 comments