The short answer
ChatGPT Dreaming is the more automated of the two — it works in the background without user input, updates entries over time, and is now expanding to free users. Claude Chat Memory is the one that reached free users first, uses a summary-based approach, and is lighter on the automation side. Both keep their memory strictly inside their own product.
What just happened
The two launches mark distinct product philosophies. Anthropic moved first on accessibility: Claude Chat Memory has been available on every plan, including free, since March 2026 (as of June 2026). It summarizes past conversations and carries context forward across sessions without requiring manual upkeep from the user.
OpenAI took longer but went further on automation. ChatGPT Dreaming, announced June 4, 2026, runs as a background process that synthesizes memory from chat history — not only from explicit facts you state, but from the full shape of your conversations. It updates entries as things change: an entry that once read "going to Singapore in July" becomes "went to Singapore in July 2026" after the trip. A memory summary page lets you review and edit what it has stored. A roughly 5x reduction in compute made it economical enough to extend to the free tier for the first time, while Plus and Pro subscribers get doubled storage. The rollout is US-first (Plus/Pro), then free and other countries to follow.
ChatGPT Dreaming vs Claude Memory at a glance
| Dimension | ChatGPT (Dreaming) | Claude Chat Memory | MemoryLake |
|---|---|---|---|
| Cross-session memory | Yes (auto-synthesized) | Yes (summarized) | Yes (verbatim) |
| Works across other AIs | No | No | Yes (any MCP tool) |
| How memory forms | Background synthesis from full history | Summary-based, lighter-touch | You supply; stored as-is |
| Updates entries over time | Yes (e.g. future→past tense) | No explicit time-update | Manual or programmatic |
| Available on free tier | Rolling out (US first, as of June 2026) | Yes, since March 2026 | Yes |
| User review/edit | Memory summary page | Settings | Fine-grained, verbatim |
| Data ownership | OpenAI-held | Anthropic-held | You own it (AES-256) |
| Version control / audit | No | No | Yes (Git-style) |
| Public accuracy benchmark | — | — | LoCoMo #1 (94.03%) |
How each one handles memory
ChatGPT Dreaming
The more automated approach. Dreaming synthesizes memory in the background from your entire chat history without you needing to tell it what to save. It tackles three real problems with earlier memory systems: staleness (entries go stale when circumstances change, and Dreaming corrects them over time), correctness (it infers context from conversations rather than relying only on stated facts), and scalability (the compute reduction that made it viable for free users). A summary page gives you visibility and control. As of June 2026, the rollout is live for Plus and Pro users in the US, with free users and other countries to follow.
Claude Chat Memory
The earlier arrival on free plans. Anthropic made Claude Chat Memory available on every subscription tier, including free, in March 2026 — predating Dreaming's free-tier reach by several months. It takes a lighter-touch approach: summarizing past conversations and carrying that context forward rather than performing aggressive automated synthesis. There is no equivalent of Dreaming's time-based entry update, but the tradeoff is a simpler, more predictable model with less background compute. Claude's memory is manageable through Settings.
What makes these two different in practice
The contrast is mostly about how much the system does automatically versus how much it relies on the shape of your stated inputs. Dreaming leans heavily on automation: it reads your full chat history, infers what to keep, and revises entries without prompting. That creates a richer, more time-aware memory but also means the system has more latitude over what it retains and how it frames it.
Claude's summary-based model is more conservative. It captures what matters from prior conversations and surfaces that context on return, but it doesn't actively rewrite prior entries as your situation changes. For users who want predictability and a clear mental model of what the assistant "knows," that lighter-touch design has its own appeal.
Neither is wrong. They reflect different bets about how much autonomy users want to cede to their memory system.
The dimension neither of them solves
Both systems do exactly one thing with the memory they build: use it in their own product. ChatGPT Dreaming's carefully synthesized picture of you lives exclusively in ChatGPT. Claude's conversation summaries stay in Claude. If you use both assistants — or add Gemini, Grok, Cursor, or any MCP-connected tool — you start with a blank slate at each one. The preferences you've built up in ChatGPT don't inform how Claude responds to you, and vice versa. You repeat yourself every time you switch.
This isn't a design flaw unique to either product. Platform memories are features for retaining users inside a product; portability would work against that purpose. But for anyone who works across multiple tools, the result is the same: fragmented context.
Where a cross-AI memory layer fits
An external memory layer takes a different approach to the problem. MemoryLake stores your context once — independently of any vendor — and exposes it over MCP, so ChatGPT, Claude, and any other MCP-connected tool can read the same source. You own the data (AES-256 encryption, exportable or deletable at any time), every change is tracked with Git-style version history, and the system scored first on the LoCoMo benchmark at 94.03% (as of June 2026). It doesn't compete with ChatGPT Dreaming or Claude Chat Memory — it adds the layer that connects what each of them keeps separately.
If you use primarily one AI, the platform memory for that tool may be enough. If you work across tools, you're effectively building a profile from scratch each time — and a portable memory layer closes that gap.
Stop rebuilding your context at every tool
ChatGPT Dreaming and Claude Chat Memory both do useful things. Use them. And let one external layer carry what you've taught each of them — so you don't start from zero every time you open a different tab.
Sources: OpenAI (ChatGPT Dreaming announcement, June 4, 2026), Anthropic (Claude Chat Memory, all-plans availability, March 2026). Feature details and rollout status current as of June 2026; verify against each vendor's latest documentation before relying on specifics.