1. GPT-5.6 Luna as the default model. Luna replaces GPT-5.5 Instant as the standard model for all non-paying users. It is the more cost-efficient member of the GPT-5.6 family, designed for speed and broad availability .
2. Unlimited text chats. OpenAI removed rate limits for text-based conversations entirely. However, this unlimited access applies only to text chats — file uploads, image generation, voice, and other tools remain subject to separate caps .
3. A new "Think" button. This optional feature lets free users trigger deeper reasoning on harder questions. When tapped, Luna spends more time processing the prompt before answering. The Think button rolled out the week after the initial announcement, subject to abuse guardrails .

Important caveat: Unlimited does not mean unrestricted. The offer is limited to text chats on Free and Go plans. OpenAI stated it would enforce abuse protections and that the feature shipped in the future tense — it was not available immediately on August 6 .

What changed for Plus and Pro users

Paying subscribers received a more substantial upgrade :

• GPT-5.6 Sol powers everything. Previously, ChatGPT for paid users routed queries between a separate "Instant" model and a "Thinking" model. Sol now handles both modes in a single, unified experience .
• A thinking slider. A new slider lets Plus and Pro users control how much reasoning effort the model applies to each response. At lower settings, Sol answers quickly; at higher settings, it reasons more deeply before replying .
• More focused, compact answers. OpenAI retuned Sol to produce more direct responses with tighter formatting and less unnecessary elaboration .

Factual accuracy: 62% and 68% fewer errors

OpenAI's headline claim for this update centers on reliability. In an internal evaluation using financial, medical, and legal prompts that demanded precise factual detail, the company found :

• GPT-5.6 Luna produced 62% fewer responses containing at least one factual error compared to GPT-5.5 Instant .
• GPT-5.6 Sol produced 68% fewer responses containing at least one factual error compared to the same baseline .

These figures come from OpenAI's own testing, not from an independent third-party benchmark . The evaluation focused on high-stakes domains — finance, medicine, and law — where factual precision matters most . The practical implication is that both models are significantly less prone to hallucination on fact-dependent queries, though users in regulated fields should still verify critical outputs against original sources .

Rollout timeline

OpenAI released the update in phases :

• Immediately (August 6): Plus and Pro users received the updated GPT-5.6 Sol with the new thinking slider .
• Same week: GPT-5.6 Luna began rolling out as the default model for Free and Go users .
• Following week (starting ~August 13): Unlimited text chats and the Think button became available to Free and Go users, subject to abuse guardrails .

OpenAI's official release notes confirmed that the changes applied specifically to the chat experience in ChatGPT, not to ChatGPT Work or Codex .

A note on the numbers

While the 62% and 68% factual-error-reduction figures dominated coverage, several analysts noted that they were internal, not independent, and limited to specific prompt types . They represent relative improvement over GPT-5.5 Instant in one evaluation set, not an absolute accuracy guarantee. For users deploying ChatGPT in production or professional workflows, these gains are encouraging but do not eliminate the need for source verification and human review in high-stakes contexts .

1. GPT-5.6 Luna as the default model. Luna replaces GPT-5.5 Instant as the standard model for all non-paying users. It is the more cost-efficient member of the GPT-5.6 family, designed for speed and broad availability .
2. Unlimited text chats. OpenAI removed rate limits for text-based conversations entirely. However, this unlimited access applies only to text chats — file uploads, image generation, voice, and other tools remain subject to separate caps .
3. A new "Think" button. This optional feature lets free users trigger deeper reasoning on harder questions. When tapped, Luna spends more time processing the prompt before answering. The Think button rolled out the week after the initial announcement, subject to abuse guardrails .

Important caveat: Unlimited does not mean unrestricted. The offer is limited to text chats on Free and Go plans. OpenAI stated it would enforce abuse protections and that the feature shipped in the future tense — it was not available immediately on August 6 .

What changed for Plus and Pro users

Paying subscribers received a more substantial upgrade :

• GPT-5.6 Sol powers everything. Previously, ChatGPT for paid users routed queries between a separate "Instant" model and a "Thinking" model. Sol now handles both modes in a single, unified experience .
• A thinking slider. A new slider lets Plus and Pro users control how much reasoning effort the model applies to each response. At lower settings, Sol answers quickly; at higher settings, it reasons more deeply before replying .
• More focused, compact answers. OpenAI retuned Sol to produce more direct responses with tighter formatting and less unnecessary elaboration .

Factual accuracy: 62% and 68% fewer errors

OpenAI's headline claim for this update centers on reliability. In an internal evaluation using financial, medical, and legal prompts that demanded precise factual detail, the company found :

• GPT-5.6 Luna produced 62% fewer responses containing at least one factual error compared to GPT-5.5 Instant .
• GPT-5.6 Sol produced 68% fewer responses containing at least one factual error compared to the same baseline .

These figures come from OpenAI's own testing, not from an independent third-party benchmark . The evaluation focused on high-stakes domains — finance, medicine, and law — where factual precision matters most . The practical implication is that both models are significantly less prone to hallucination on fact-dependent queries, though users in regulated fields should still verify critical outputs against original sources .

Rollout timeline

OpenAI released the update in phases :

• Immediately (August 6): Plus and Pro users received the updated GPT-5.6 Sol with the new thinking slider .
• Same week: GPT-5.6 Luna began rolling out as the default model for Free and Go users .
• Following week (starting ~August 13): Unlimited text chats and the Think button became available to Free and Go users, subject to abuse guardrails .

OpenAI's official release notes confirmed that the changes applied specifically to the chat experience in ChatGPT, not to ChatGPT Work or Codex .

A note on the numbers

While the 62% and 68% factual-error-reduction figures dominated coverage, several analysts noted that they were internal, not independent, and limited to specific prompt types . They represent relative improvement over GPT-5.5 Instant in one evaluation set, not an absolute accuracy guarantee. For users deploying ChatGPT in production or professional workflows, these gains are encouraging but do not eliminate the need for source verification and human review in high-stakes contexts .