What changes did OpenAI announce on August 7, 2026, regarding GPT-5.6 Luna becoming the default model for Free and Go tier ChatGPT users, including unlimited text chats, a new "Think" button, safety guardrails, factual accuracy improvements over GPT-5.5 Instant, and how do these

On August 6–7, 2026, OpenAI announced a major restructuring of its ChatGPT model tiers: GPT-5.6 Luna became the default for Free and Go users with unlimited text chats, GPT-5.6 Sol was retuned for paid subscribers with a new reasoning slider, and the company cited competitive pricing pressure from Chinese open-weight models behind Luna's 80% API price cut .

Changes for Free and Go Users (GPT-5.6 Luna)

• Default model swapped to GPT-5.6 Luna, replacing GPT-5.5 Instant. Luna is OpenAI's fastest and most affordable model .
• Unlimited text chats for all Free and Go users — the previous rate limits on text conversations are removed entirely .
• New "Think" button lets free users escalate reasoning on a per-message basis for harder questions that need more thought. This is described as a "per-message reasoning escalation" compared to the continuous slider on paid tiers .
• Abuse guardrails apply to the unlimited chat expansion to prevent misuse .
• Factual accuracy improvement: GPT-5.6 Luna makes approximately 62% fewer factual errors compared to GPT-5.5 Instant, according to OpenAI's internal evaluations .
• Rollout is staged: Luna as default model rolled out the week of August 3, unlimited text chats and the Think button the week of August 10 .

Changes for Paid Subscribers (GPT-5.6 Sol)

• GPT-5.6 Sol retuned for ChatGPT: now produces more concise, direct, and fact-focused answers, tuned for everyday chat use rather than the original agentic/research-heavy profile .
• ~68% fewer factual errors than GPT-5.5 Instant in OpenAI's internal testing on financial, medical, and legal prompts — answers containing at least one factual mistake dropped significantly .
• New reasoning slider (continuous effort control) lets Plus and Pro users dial up or down how much thought Sol puts into each response, replacing the older fixed reasoning levels .
• Sol remains exclusive to Plus, Pro, Business, and Enterprise — it is not available on free plans .

Luna vs. Sol: Key Differences

Dimension	GPT-5.6 Luna (Free/Go)	GPT-5.6 Sol (Paid)
Reasoning control	Per-message "Think" button (binary on/off)	Continuous reasoning slider
Best for	Speed, everyday questions, affordability	Complex agentic work, coding, research
API price (per 1M tokens)	$0.20 input / $1.20 output	$5 input / $30 output
Factual accuracy vs GPT-5.5	~62% fewer errors	~68% fewer errors
Availability	Free, Go, and paid plans	Plus, Pro, Business, Enterprise only

Competitive Pressure and the 80% API Price Cut

• Luna's API price was slashed 80% on July 30, 2026 — from $1/$6 per 1M tokens down to $0.20/$1.20 — with Terra cut 20% as well. OpenAI explicitly framed this as responding to the "price-performance frontier" .
• Chinese open-weight models are the primary pressure. DeepSeek's V4 Flash is roughly 105 times cheaper to run than Anthropic's Claude Fable 5 on benchmark tests, and DeepSeek V4 Pro prices output at 1/57th the cost of Fable 5 .
• Alibaba's Qwen 3.8-Max (2.4 trillion parameters, open-weight) launched just days earlier, matching top Western models at a fraction of the cost .
• By mid-2026, Chinese open-weight models accounted for ~61% of all tokens consumed on OpenRouter, the largest neutral AI inference platform .
• The Los Angeles Times described this as a "death zone" for US AI companies without frontier tech or market-breaking pricing — the Luna price cut and free tier expansion are OpenAI's direct response .