Muse Wonder Remembers You. Now Make Claude, ChatGPT and Codex Remember Too.

Yes, Muse Wonder remembers between conversations. Meta launched Muse Wonder and Muse Wonder Full on October 12, 2026, and the headline feature is Active Memory, which carries important information across conversations and pulls it back in when it becomes relevant. That memory lives inside Meta's products, though, so if you also use Claude, ChatGPT or Codex, they still start from zero unless you give all of them one shared memory layer like Vilix AI.

What did Meta launch?

Muse Wonder is a family of two frontier models from Meta Superintelligence Labs, built for agents that do personal and professional work on your behalf:

• Muse Wonder: the fast version, available now through the Meta Model API and Muse Code.
• Muse Wonder Full: the full version, limited at launch to Muse Code's higher subscription plans.

Both support a 1 million token context window. API pricing reported at launch is about $4 per million input tokens and $20 per million output tokens. Meta says Wonder will soon power the subscription version of the Muse personal agent app, with plans reported at $20 or $100 a month.

Two features stand out. Active Memory carries context across conversations. Active Focus is meant to keep the model working when a task gets hard, try other approaches, and recover from errors.

How good is Muse Wonder's memory?

On Meta's own MemoryBench evaluation, which measures how well a model remembers and uses context over a long interaction history, the scores at maximum effort were:

Keep in mind this is Meta's internal benchmark, and independent testing is still coming in. But the bigger signal is that memory is now something frontier labs compete on and publish scores for. A year ago it was a side feature. Now it's a headline.

Does one model win everything?

No, and that's the most useful part of Meta's own scorecard. Look at who leads which row:

• Muse Wonder Full leads legal agent work (Harvey LAB), finance research (Vals Finance Agent), business workflows (AutomationBench) and long-horizon coding (DeepSWE, SWE-Marathon).
• Claude Opus 5.5 still leads agentic terminal coding (Terminal-Bench), from-scratch programming (ProgramBench), professional tool use (JobBench) and the overall Artificial Analysis Index.
• GPT-6 Astra leads professional document work (GDP.pdf).

So the smart setup in late 2026 isn't picking one model. It's using the best one for each job: plan on your phone in ChatGPT, build in Claude Code or Codex, hand legal and finance research to an agent running Muse Wonder. And with Meta pricing Wonder below the leading models, switching between them only gets more common.

What's the catch with built-in memory?

Every lab's memory stays inside its own product. Muse's Active Memory knows what you told Muse. Claude's memory knows what you told Claude. ChatGPT's memory knows what you told ChatGPT. None of them read each other.

The moment you split work across models, you're back to re-briefing: the project, the decisions, the client's tone, what already failed. The better each model's memory gets, the more you notice that it doesn't follow you to the next tool.

How do you share one memory across Muse, Claude, ChatGPT and Codex?

Vilix AI is one shared memory for all your AI tools. You connect each tool to one Vilix AI account once, and they all share the same context, memory, projects, rules and tasks.

• Cloud, not local. Vilix AI is a remote MCP server, cloud-hosted and fully managed. There's nothing to install, host or maintain.
• Follows you everywhere. Because it's MCP in the cloud, the same memory works on phone apps (Claude, ChatGPT), the desktop browser, desktop apps (ChatGPT, Codex, Claude, Claude Code, Claude Cowork, Cursor) and terminal agents (Claude Code CLI, Codex CLI).
• Full history, not just facts. It keeps full conversation history with no limit and pulls back only what's relevant.
• Work state, not just chat. Projects and tasks, personal rules and project rules, and agent skills: build a procedure once and every agent follows it. When two tools disagree, the newest write wins.
• Your data stays yours. Export your whole memory anytime, or delete it anytime.

In practice, each AI checks Vilix AI at the start of a session and saves what happened at the end. Plan a feature in ChatGPT on your phone, open Claude Code or Codex on your laptop, and it already knows the project. Muse is on the supported list too, so the context you build in Muse can sit next to everything else.

It works with ChatGPT, Claude, Claude Code, Codex and Cursor, plus OpenClaw, Hermes, Gemini Spark, Gemini CLI, Grok, Grok Build, Manus, GitHub Copilot, Windsurf, Devin, Lovable, Cline, Continue, Muse and any MCP-compatible AI. Plan and permission requirements of each tool apply.

One honest tradeoff: Muse's Active Memory works the second you start chatting in Muse, with zero setup. Vilix AI asks you to connect each tool once. That one-time step is what lets the memory travel between models.

Who should care most?

AI agencies and automation builders. Give each client their own Vilix AI account and wire every agent you build for them to that client's key. When a new model like Muse Wonder wins a task you run for a client, you swap the model and the client's whole history comes along.

Solo founders and indie devs. If you pay for two or three AI subscriptions, you've already felt the re-briefing tax. One memory layer ends it.

Watch: Liam and Stella break this down on Agentic Pod, with the full technical guide on GitHub.

FAQ

Does Muse Wonder remember between conversations?

Yes. Active Memory carries important information across conversations and brings it back when it's relevant.

Can Claude or ChatGPT see what Muse remembers?

Not on their own. Each lab's built-in memory stays in its own product. A shared memory layer over MCP, like Vilix AI, is how different models read the same context.

Where can I use Muse Wonder today?

Through the Meta Model API and Muse Code. Meta says it will roll out to Muse app subscribers soon.

Is Vilix AI free?

There's a free plan forever with no credit card, a 7-day Pro trial with no credit card, and paid Pro and Power plans.

Related: Meta Just Launched Muse: Does It Actually Remember You? and Muse + Vilix AI: Every AI I Use Now Shares One Memory.

Try it: connect your first tool at vilix.ai.