llms.blog
OpenAI

OpenAI Improves GPT-5.6 Sol for Paying Users, Moves Free Tier to Luna Only

Paying ChatGPT users get a more focused Sol model with a reasoning effort slider. Free users gain unlimited text chats but lose access to OpenAI's strongest reasoning model.

axc
axc
2 min
OpenAI Improves GPT-5.6 Sol for Paying Users, Moves Free Tier to Luna Only

OpenAI has updated GPT-5.6 Sol in ChatGPT for Plus and Pro subscribers, shipping a version that the company says cuts unnecessary detail and excess formatting. At the same time, free-tier users are being shifted to GPT-5.6 Luna, the smallest and cheapest model in the GPT-5.6 family, with unlimited text chats set to arrive next week.

What changed in Sol

The Sol update brings two changes for paying users. The model now produces shorter, more direct answers for simple queries while preserving depth for complex tasks. OpenAI also claims a reduction in factual errors. In an internal evaluation using prompts from finance, medicine, and law, responses containing at least one factual mistake dropped by roughly 62 percent for Luna and 68 percent for Sol compared with GPT-5.5 Instant. These figures have not been independently verified.

A new reasoning-effort slider lets paying users choose from five levels of processing depth. Lower settings handle everyday questions, while higher settings are intended for research, planning, and coding. The slider was previously available only in ChatGPT Work. OpenAI positions it as a way to make quick answers and deep reasoning feel like one model rather than two separate experiences.

OpenAI's two-tier model access: Sol for paying users, Luna for free tier

The free tier tradeoff

GPT-5.6 Luna becomes the default model for Free and Go users later this week. Unlimited text chats follow next week, along with a Think button that lets Luna reason longer on harder questions. But Luna does not switch to a stronger model when it struggles. Smaller models are generally more error-prone than larger reasoning counterparts, and the Think button extends Luna's processing time without upgrading the underlying model. Free users lose access to OpenAI's most capable reasoning altogether.

The Sol changes apply only to ChatGPT. The model remains unchanged in ChatGPT Work and Codex. Limits on file uploads, image generation, and other tools stay in place for free users.

The update is the latest step in a tiered strategy that has defined OpenAI's 2026 product releases. The company cut GPT-5.6 Luna API pricing by 80 percent shortly after launch and has since used Luna as the entry point for free users while reserving Sol for paying subscribers. Whether a reasoning slider meaningfully improves the experience remains an open question. OpenAI's own model switcher, introduced with GPT-5, was largely ignored by users.

Sources

OpenAI: Improving GPT-5.6 Sol in ChatGPT: https://openai.com/index/improving-gpt-5-6-sol-in-chatgpt/OpenAI ChatGPT Release Notes: https://help.openai.com/en/articles/6825453-chatgpt-release-notesThe Decoder: OpenAI improves GPT-5.6 Sol in ChatGPT and restricts free users to its weakest model: https://the-decoder.com/openai-improves-gpt-5-6-sol-in-chatgpt-and-restricts-free-users-to-its-weakest-model/
OpenAIChatGPTModels
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Decentralized and Peer-to-Peer LLM Inference in Production: Architecture, Ring Memory Partitioning, and Network Latency

    Decentralized and Peer-to-Peer LLM Inference in Production: Architecture, Ring Memory Partitioning, and Network Latency Frontier open-weight models such as Llama 3.1 405B, DeepSeek-V3, and Command R+ have expanded model capabilities, but their parameter scales exceed the physical memory limits of individual consumer and edge workstations. Running a 405-billion parameter model in 16-bit precision requires over 810 GB of memory, and even 4-bit quantized variants require roughly 230 GB of contiguo

    1 min

  • Discrete Diffusion in Large Language Models: How Continuous-Time Markov Chains, Absorbing States, and Score Entropy Challenge Autoregressive Generation

    The dominance of autoregressive architectures in large language models rests on a fundamental mathematical formulation: the chain rule of probability. By factoring the joint distribution of a sequence into a product of conditional probabilities, $p(x) = \prod_{i=1}^N p(x_i \mid x_{<i})$, autoregressive models reduce text generation to sequential next-token prediction. While this left-to-right causal factorization has scaled effectively across compute regimes, it imposes rigid operational constr

    1 min

  • AI Agents Surpass Humans on OpenRouter as Agentic Token Usage Jumps 14x

    Autonomous AI agents have overtaken human users as the primary consumers of language model compute on OpenRouter, with agentic token volume surging fourteenfold over the past six months. Data published by OpenRouter analyst Peter Walker indicates that February 6 marked the permanent inflection point where token consumption by automated agents exceeded direct human API traffic. Since that threshold, agentic token volume on the multi-model gateway has climbed from 0.51 trillion to 7.3 trillion to

    1 min