OpenAI updated ChatGPT today, tuning GPT-5.6 Sol for paid users and adding a reasoning effort slider, while making GPT-5.6 Luna the default free model with unlimited text conversations and a new Think button. The update focuses on improving factual accuracy and reducing inference costs, enabling unlimited free chat.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
OpenAI has set GPT-5.6 Luna as the default model for the free tier and opened unlimited text conversations, driven by an 80% API price cut and targeted tuning for conversations. This article details the update specifics, internal evaluation numbers, and safety measures, while noting the vendor's self-assessment and applicability limits.
On August 7, OpenAI updated ChatGPT. On the paid side, the Chat version of GPT-5.6 Sol for Plus and Pro users has been retuned, adding a slider to control 'thinking effort'. On the free side, the default model is now GPT-5.6 Luna, text conversations are no longer limited, and a Think button has been added.
The tuning of Sol is limited to the Chat experience; the Sol versions powering Work and Codex remain unchanged, representing a targeted optimization for everyday conversations. The tuning direction is 'convergence': adjusting detail based on the question, avoiding formatting overkill, and proactively correcting when agreement is unhelpful, responding to criticism of sycophancy.
OpenAI emphasizes that Instant and deep reasoning are driven by the same model; increasing the thinking level means 'the same model thinks longer', not switching models. The slider is now available to Plus and Pro users on web, mobile, and desktop.
OpenAI provides factual accuracy data: on a set of prompts involving finance, healthcare, and law requiring specific facts, the proportion of responses containing at least one factual error is about 62% lower for GPT-5.6 Luna compared to GPT-5.5 Instant, and about 68% lower for Sol.
It should be noted that these are OpenAI's internal evaluations; the number of questions, scoring methods, and error criteria are not disclosed, representing a vendor self-assessment that cannot be independently verified externally.
GPT-5.6 launched on July 9, with three tiers: Sol, Terra, and Luna. On July 30, OpenAI cut API prices: Luna by 80% to $0.2 per million input tokens and $1.2 per million output tokens; Terra by 20%; Sol unchanged.
Previously, Luna and Terra were used only for Work, Codex, and API, with Luna described as an 'efficiency tier' for high-throughput, low-latency tasks. This is the first time it becomes the default model for everyday conversations for regular users.
Free unlimited is less a gesture than a natural step after unit cost decline. Reduced inference costs directly translate into the service form of the free tier.
Unlimited usage applies only to text conversations; file uploads, images, and other tools still have quotas and are subject to abuse protections. The Think button for free and Go users calls Luna, not Sol; they are different.
GPT-5.6 Luna becomes the default model this week, with unlimited conversations and the Think button rolling out next week.
On safety, the system card mentions additional training for suspected minors, including avoiding romantic role-play and age-restricted challenges, not presenting itself as a substitute for real-life relationships, and encouraging teens to contact trusted people when needed. OpenAI states it has added corresponding evaluations for users under 18.
This article is based on Machine Intelligence's report on OpenAI's official update. Core facts (model names, prices, timeline) come from official sources, but factual accuracy numbers are vendor self-assessments without disclosed evaluation details; readers should treat them with caution.
ChatGPT's free unlimited offering results from cost decline and model positioning adjustments, but note its text-only limitation and vendor self-assessment.
Primary report
Primary source