OpenAI's GPT-5.6 Shake-Up: Free Users Get Luna, Paid Users Get a Smarter Sol

GPT-5.6 Luna Goes Free: Unlimited Chats and a Think Button

The headline change for the roughly one billion weekly ChatGPT users is that GPT-5.6 Luna is now the default model for all free-tier and Go ($8/month) plan users, replacing GPT-5.5 Instant .

• Unlimited text chats: OpenAI has removed rate limits on text conversations entirely for free and Go users . Limits on file uploads, images, voice, and image generation remain in place.
• New 'Think' button: This is a per-message escalation tool, allowing free users to allocate more reasoning power to difficult questions. It's a binary on/off switch, contrasting with the continuous slider on paid plans .
• Factual accuracy improvement: According to OpenAI's internal evaluations on financial, medical, and legal prompts, GPT-5.6 Luna makes approximately 62% fewer factual errors than GPT-5.5 Instant .
• Staged rollout: The model default change happened the week of August 3, 2026. Unlimited text chats and the Think button began rolling out the following week .
• Abuse guardrails were implemented alongside the unlimited chat expansion .

Note: The unlimited text chat feature is specific to the main ChatGPT chat experience. It does not apply to ChatGPT Work or Codex, which remain unchanged in this update .

GPT-5.6 Sol: Retuned for Paid Users

For Plus, Pro, Business, and Enterprise subscribers, the flagship GPT-5.6 Sol model has been significantly retuned .

• More concise and direct: Sol is now tuned for everyday conversation, producing tighter, more fact-focused responses instead of the original agentic/research-heavy style .
• Dramatic accuracy gains: Internal testing on financial, medical, and legal prompts showed responses containing 68% fewer factual errors compared to GPT-5.5 Instant .
• New reasoning slider: This replaces fixed reasoning levels, giving users continuous control over how much effort Sol puts into each response .

Sol remains exclusive to paid plans. It is not available to free or Go users .

Luna vs. Sol: A Side-by-Side Comparison

The two models now serve very different use cases and user segments.

The Driving Force: Chinese Open-Weight Models

Behind these user-facing changes is a fierce pricing war, driven by the rapid rise of Chinese open-weight AI models .

• The 80% price cut: On July 30, 2026, OpenAI slashed GPT-5.6 Luna's API price by 80% (from $1/$6 to $0.20/$1.20 per 1M tokens) and cut Terra's price by 20%. The company explicitly framed this as responding to the "price-performance frontier" .
• The competition is cheap: DeepSeek's V4 Flash model is estimated to be 105 times cheaper to run than Anthropic's Claude Fable 5 on benchmark tests, with V4 Pro priced at 1/57th the cost .
• Alibaba's challenge: Alibaba launched Qwen 3.8-Max, a 2.4 trillion-parameter open-weight model, just days before the OpenAI announcement. It is reported to match top Western models at a fraction of the cost .
• Market share shift: By mid-2026, Chinese open-weight models accounted for ~61% of all tokens consumed on OpenRouter, the largest neutral AI inference platform .
• A 'death zone' for some: The Los Angeles Times characterized the environment as a "death zone" for US AI companies that lack either true frontier technology or market-breaking pricing .

The message is clear: OpenAI is cutting prices and expanding access to compete not just with other US labs, but with a flood of capable, low-cost open-weight models from China.