Which new models are on Secret Chat AI?

GPT 6 Sol, GPT 6 Luna, Claude Opus 5.5 and Grok 4.7 are live. Grok 4.7 arrived on September 22, 2026. The GPT 6 and Claude Opus 5.5 changes went live on September 23.

What is new in GPT 6 Sol and Luna?

OpenAI says both models bring advances from GPT 6 Astra into faster models and improve on their GPT-5.6 counterparts. Sol is intended for complex work such as coding, while Luna handles clear, high-volume tasks such as summaries, extraction and quick answers.

Why does ChatGPT Web Search use GPT 6 Sol?

OpenAI has not released a GPT 6 Terra. Until a middle GPT 6 model appears, the more capable GPT 6 Sol covers Web Search as well as Research.

What is new in Claude Opus 5.5?

Anthropic says Opus 5.5 is more than 30 percent faster than Opus 5, writes more naturally, reads dense visual material more accurately and is built for long-running coding and knowledge work. It supports a one-million-token context window and outputs up to 128,000 tokens.

Does Claude Research still use high effort?

Yes. Opus 5.5 defaults to medium effort, but Secret Chat explicitly requests high effort for Opus. Claude Research therefore keeps the deeper setting.

What changed in Grok 4.7?

xAI says Grok 4.7 uses a new, larger base model, works longer on difficult tasks, checks its work more carefully and handles longer context better. Its text and image inputs and reasoning-effort levels remain the same.

Do old chats still work after the model changes?

Yes. Open an existing ChatGPT, Claude or Grok conversation and continue normally. The next reply uses the current model assigned to that mode.

Are the new models available in the AI Council?

GPT 6 and Grok 4.7 are. The ChatGPT seat runs GPT 6 Luna or GPT 6 Sol through Web Search on the heavier rung, Grok's seat runs Grok 4.7, and GPT 6 Luna is the referee. The Claude seat remains on Haiku 4.5 or Sonnet 5, so Claude Opus 5.5 is not in the council.

Is access to these models anonymous?

Secret Chat does not tie your request to a personal OpenAI, Anthropic or xAI account, builds no profile of you, stores no server-side chat archive and never uses prompts for training. Your prompt is sent verbatim to the provider, so remove identifying or sensitive details before sending.

GPT 6, Claude Opus 5.5 and Grok 4.7 on Secret Chat AI

Published September 23, 2026

GPT 6 Sol, GPT 6 Luna, Claude Opus 5.5 and Grok 4.7 are live on Secret Chat AI. Grok 4.7 arrived here on September 22. The two GPT 6 models and Claude Opus 5.5 followed on September 23. There is nothing to install or configure, and old chats continue on the new model assigned to that mode.

The useful question is what changed inside each model and where you will meet it in the app. The answer is different for each family: OpenAI split fast, high-volume work from harder coding and reasoning; Anthropic made its Opus tier faster and stronger on long, demanding work; and xAI gave Grok a larger base model with more time to check difficult answers.

GPT 6 Sol and Luna: Two Models for Different Work

OpenAI announced GPT 6 Sol and GPT 6 Luna on September 22. OpenAI says the pair brings much of the progress behind GPT 6 Astra into faster models that can support work at scale. It also says both models build on Astra's alignment advances and improve on their GPT-5.6 counterparts.

The split is practical. According to OpenAI's description reported by TechCrunch, Sol is designed for complex tasks such as coding. Luna is aimed at high-volume work with a clear goal: summarizing documents, extracting information and answering quick questions. OpenAI says Sol made roughly 50 percent fewer mistakes than its predecessor in an internal factuality evaluation built from real-world conversations, and that its coding error rate was lower. Those are OpenAI's evaluations, not tests run by Secret Chat.

Where GPT 6 Sol and Luna Run on Secret Chat

Fast Chat runs on GPT 6 Luna.
Smart Agentic uses Luna as its router and fast tier, then escalates hard questions to GPT 6 Sol.
Research runs on GPT 6 Sol.
Web Search runs on GPT 6 Sol for now. OpenAI has not released a middle GPT 6 model called Terra, so the more capable model covers that slot until one appears. Gemini uses a similar arrangement: Web Search and Research share Gemini 3.8 Flash because no Gemini Pro is available to us.
Perplexity Search uses GPT 6 Luna through Perplexity's Agent API.
ChatGPT titles and conversation summaries are written by GPT 6 Luna.

GPT 6 Luna also occupies the ChatGPT seat in the free DeepSeek + ChatGPT council pair and serves as the AI Council referee. The separate GPT 6 Astra mode has not changed.

Claude Opus 5.5: Faster, Clearer and Built for Long Work

Anthropic released Claude Opus 5.5 on September 22 and calls it a major step up from Opus 5. According to Anthropic, early testers reported large gains on their most complex work. The company says the new model produces output more than 30 percent faster than Opus 5 and communicates more naturally, with writing that testers found clearer and easier to follow.

Anthropic's documentation says Opus 5.5 is much sharper at reading values from dense charts, diagrams and screenshots. It thinks more per turn at a given effort setting, keeps adaptive thinking on, and is built for long-running agentic coding and knowledge work. Its context window holds one million tokens, with support for outputs up to 128,000 tokens. In Anthropic's automated behavioral audit, Opus 5.5 was the strongest-performing model the company had tested at publication. These are Anthropic's findings.

Where Claude Opus 5.5 Runs on Secret Chat

Claude Opus 5.5 now runs Claude Research and the tier that Smart Agentic Claude selects for hard questions. Haiku 4.5 still handles Fast Chat, routing, titles and summaries. Sonnet 5 remains on Web Search, while Claude Fable 5.1 remains a separate mode you select yourself.

Anthropic changed the default effort from high on Opus 5 to medium on Opus 5.5. Secret Chat explicitly sends high effort for Opus, so Research keeps working at the deeper setting. You do not need to change anything.

Grok 4.7: A Larger Base Model That Checks Its Work

xAI released Grok 4.7 on September 21. The company says it works longer on difficult tasks and checks its work more carefully. Underneath, it uses a new, larger base model than Grok 4.6 and was trained with a longer reinforcement-learning run on a harder mix of tasks.

xAI also says Grok 4.7 is better at verifying its own work and managing longer context, and describes its safeguards as the company's best calibrated so far. Text and image input remain as before, as do the available reasoning-effort levels.

Where Grok 4.7 Runs on Secret Chat

Grok 4.7 now powers every Grok text tier:

Smart Agentic and its router;
Fast Chat, with reasoning turned off;
Web Search, with xAI's web and X search tools;
Research, with high reasoning and search;
chat names and conversation summaries.

Grok's image generator is a separate model and did not change. Before switching the text tiers, we checked Grok 4.7 with a live call and deleted the stored response afterwards.

The AI Council Lineup

The AI Council sends one question to several model families and asks a referee to compare their answers. The ChatGPT seat now uses GPT 6 Luna, or GPT 6 Sol through Web Search on the heavier rung. Grok's seat uses Grok 4.7. GPT 6 Luna is the referee.

The Claude seat remains on Haiku 4.5 or Sonnet 5, so Claude Opus 5.5 is not part of the council. The panel is built to expose agreement and disagreement between companies; to get an answer from Opus 5.5, use Claude Research or let Smart Agentic Claude escalate.

What You See in the App

The controls have not moved. Choose ChatGPT, Claude or Grok, leave Smart Agentic on, or select a mode yourself. Under every reply, the app names the model that wrote it: GPT 6 Sol, GPT 6 Luna, Claude Opus 5.5 or Grok 4.7.

Existing conversations continue on the replacement model. Where the provider lets us, we delete its stored copy of a response after you get the answer. Answers written by GPT-5.6 or Claude Opus 5 before the switch are still found and deleted the same way, and the Session Privacy report lists them.

How Anonymous Access Works

Secret Chat forwards each prompt to OpenAI, Anthropic or xAI without tying it to a personal account at that provider. We build no profile of you, store no archive of your chats on our servers, and prompts are never used for training. Your chat history stays in your browser.

The limit matters: your prompt goes to the provider verbatim so it can answer. Anonymity removes the account link; it does not remove names, case numbers or other identifying details you type into the message. Redact sensitive details before sending when the words themselves could identify you.

Frequently Asked Questions

Which new models are on Secret Chat AI?
GPT 6 Sol, GPT 6 Luna, Claude Opus 5.5 and Grok 4.7 are live. Grok 4.7 arrived on September 22, 2026. The GPT 6 and Claude Opus 5.5 changes went live on September 23.
What is new in GPT 6 Sol and Luna?
OpenAI says both models bring advances from GPT 6 Astra into faster models and improve on their GPT-5.6 counterparts. Sol is intended for complex work such as coding, while Luna handles clear, high-volume tasks such as summaries, extraction and quick answers.
Why does ChatGPT Web Search use GPT 6 Sol?
OpenAI has not released a GPT 6 Terra. Until a middle GPT 6 model appears, the more capable GPT 6 Sol covers Web Search as well as Research.
What is new in Claude Opus 5.5?
Anthropic says Opus 5.5 is more than 30 percent faster than Opus 5, writes more naturally, reads dense visual material more accurately and is built for long-running coding and knowledge work. It supports a one-million-token context window and outputs up to 128,000 tokens.
Does Claude Research still use high effort?
Yes. Opus 5.5 defaults to medium effort, but Secret Chat explicitly requests high effort for Opus. Claude Research therefore keeps the deeper setting.
What changed in Grok 4.7?
xAI says Grok 4.7 uses a new, larger base model, works longer on difficult tasks, checks its work more carefully and handles longer context better. Its text and image inputs and reasoning-effort levels remain the same.
Do old chats still work after the model changes?
Yes. Open an existing ChatGPT, Claude or Grok conversation and continue normally. The next reply uses the current model assigned to that mode.
Are the new models available in the AI Council?
GPT 6 and Grok 4.7 are. The ChatGPT seat runs GPT 6 Luna or GPT 6 Sol through Web Search on the heavier rung, Grok's seat runs Grok 4.7, and GPT 6 Luna is the referee. The Claude seat remains on Haiku 4.5 or Sonnet 5, so Claude Opus 5.5 is not in the council.
Is access to these models anonymous?
Secret Chat does not tie your request to a personal OpenAI, Anthropic or xAI account, builds no profile of you, stores no server-side chat archive and never uses prompts for training. Your prompt is sent verbatim to the provider, so remove identifying or sensitive details before sending.

GPT 6, Claude Opus 5.5 and Grok 4.7 Are Here:
Three Model Families Changed on Secret Chat AI