← Back to feed

OpenAI has moved GPT-5.6 Luna into the free tier of ChatGPT as the default model, replacing the older model that free users had been on since the 5.5 Instant rollout in May. Free accounts now get unlimited text chats. A new Think button lets free users escalate to higher-reasoning mode for harder questions on demand.

For paid users, Sol is getting a ChatGPT-specific tuning pass distinct from its API version. The changes target factual reliability and response formatting. In internal evaluations across financial, medical, and legal prompts that required precise factual detail, responses containing at least one error were 68% less common with the updated Sol than with GPT-5.5 Instant. Luna on the same benchmark came in at 62% fewer errors.

A new slider lets Plus and Pro users set how much reasoning Sol applies to each response, creating a single model that covers both instant answers and extended thinking without switching between tools.

What the Free Tier Upgrade Means

GPT-5.6 Luna on free accounts is not the same as Luna on the API. The consumer product is separately tuned for conversation quality — more direct answers, better calibration between question complexity and response length, and tighter formatting. OpenAI’s product note specifically calls out avoiding unnecessary formatting on simple questions and providing fuller responses for multi-step tasks.

The unlimited text chat point matters more than it sounds. Prior free-tier limits pushed users toward session resets or workarounds. Removing that cap with a 5.6-generation model is a distribution move: OpenAI is now fielding one of its two newest production models at no cost, narrowing the gap between free and paid that had been widening since Sol launched under government gate in late June.

Sol, Tuned for 1 Billion Weekly Users

OpenAI’s framing for the Sol update is explicitly about at-scale use patterns. With 1 billion weekly users, the range of tasks hitting ChatGPT spans quick lookup to multi-step research. The tuned Sol attempts to adapt response length and detail to the actual demand of the query rather than applying a uniform style.

The factual accuracy work targets the specific error modes that appear on fact-dependent prompts: dates, numbers, source attribution, legal rules, medical dosing. Internal testing covered financial, medical, and legal domains — exactly the categories where confabulation risk is highest and where GPT-5.5 Instant has drawn the most criticism.

The reasoning slider consolidates what had previously required switching between Instant and Thinking modes. From August 7, Plus and Pro users get a single Sol interface with variable reasoning depth.

Competitive Context

Luna as a free-tier default is a meaningful shift in the access landscape. Gemini’s free tier runs Gemini 2.5 Flash. Claude.ai free runs Claude Sonnet 5. GPT-5.6 Luna at a similar tier is OpenAI’s response — placing a frontier-class model in front of users who were previously several generations behind paying users.

The 62%/68% factual error reduction figures are internal evaluations, not third-party verified. But they are specific to financial, medical, and legal use cases, which suggests OpenAI is targeting enterprise-adjacent use cases even in the consumer product.

Luna’s API price was cut 80% in late July to $0.30/M input, $1.20/M output. The free-tier rollout follows a pattern: price cut to drive developer adoption, then consumer expansion to drive volume.

Key Numbers

• GPT-5.6 Luna: new default for free ChatGPT users, unlimited text chats
• GPT-5.6 Sol (ChatGPT tuned): 68% fewer factual errors vs GPT-5.5 Instant on financial, medical, legal prompts
• GPT-5.6 Luna (ChatGPT tuned): 62% fewer factual errors vs GPT-5.5 Instant on same benchmark
• New: Think button for free users (on-demand higher reasoning)
• New: Reasoning-depth slider for Sol (Plus/Pro)
• Luna API pricing (from July 30 cut): $0.30/M input, $1.20/M output