Petals vs ChatGPT memory

ChatGPT is a genuinely impressive product. It answers questions well, generates images with DALL-E, produces video with Sora, holds real-time voice conversations with low latency, and now — with Dreaming V3, launched June 2026 — maintains a memory of your life that updates itself in the background without you lifting a finger. For a lot of people, it is the right tool. That is worth saying before anything else.

How ChatGPT remembers

ChatGPT has two memory layers running side by side.

The original layer, still present, stores specific facts as a flat list of text snippets — your name, preferences, goals, how you like to be addressed. It can hold roughly 1,200–1,400 words before old entries must go to make room for new ones. You have always been able to add, correct, or dismiss entries from Settings → Personalization → Manage Memory.

The new layer is Dreaming V3. It runs as a background process that reads across your entire conversation history and rewrites a memory state autonomously. It captures context that arises naturally — "you're going to Singapore in July" — and revises it as time passes. By OpenAI's own numbers, factual recall improved from 41.5% at the 2024 baseline to 82.8% with Dreaming V3. That is a meaningful jump.

The trade-off is visibility. OpenAI acknowledges that the Memory Summary Page "does not necessarily include everything ChatGPT may remember." Dreaming V3 synthesizes inferences from years of conversation history, and not all of them surface in the list you can see. You can edit what you can see — but the system is drawing conclusions you never explicitly confirmed, and there is no complete audit trail for them.

One additional data point worth knowing: Tenable Research (November 2025) demonstrated that memories appended to system prompts can be exploited via third-party sources to create a persistent exfiltration channel across sessions. OpenAI has been working on mitigations, but the structural risk of opaque memory persisting in your system prompt is real.

How Petals remembers

Petals builds a knowledge graph — a network of entities (people, places, projects, ideas), relationships between them, and the claims that define each node. Every claim carries source traceability: you can open any node and see which conversation it came from.

The graph is not a summary you read once and move on. It is something you browse, navigate, and edit. You can correct a node, delete a relationship, or prune a claim that is no longer true. When new information contradicts something already in the graph, the system surfaces the conflict rather than silently overwriting it.

Ingestion today is manual: paste notes, upload documents, drop in links or files, share transcripts. The ingestion API exists for people who want to wire up automated feeds — a script that pulls from Screenpipe, Oura, or Boox handwritten notes and posts to the transcript endpoint, authenticated with an API key from Settings → API keys. That is a bring-your-own-script workflow today, not a one-click connector. The landing page calls this the "adventurous" path, which is fair.

Petals does not synthesize inferences in the background and hope you notice the results later. What is in the graph is what you put there — directly or via conversations — and what the AI extracted with your knowledge.

Where ChatGPT is better

This section deserves honesty, so here it is plainly.

Multimodality. ChatGPT generates images with DALL-E, video with Sora, and holds real-time voice conversations on mobile via Advanced Voice Mode. It accepts screen and video input on Plus. Petals is text-first. If image generation or low-latency voice is part of your daily use, ChatGPT is the stronger choice today.

Model breadth and ecosystem. GPT-5.5 Pro is at benchmark parity with the best Claude models on most tasks and is meaningfully more token-efficient on high-volume agentic work. ChatGPT integrates with Microsoft 365 via Copilot. It has a large marketplace of custom GPTs with specialized knowledge. It has ChatGPT Tasks for scheduled recurring interactions. The surface area is larger.

Automatic temporal updating. Dreaming V3 revises stale facts without you doing anything. "You went to Singapore in July 2026" replaces "you're going to Singapore in July" automatically. If you do not want to maintain a memory profile manually, this is a genuine convenience win.

Name recognition and install base. ChatGPT has a vastly larger user base, stronger community resources, more tutorials, and broader third-party tooling. If you want ecosystem depth, it is there.

Pricing at the free tier. ChatGPT's free tier includes basic memory, with Dreaming V3 rolling out to free users in the weeks following the June 2026 Plus launch. Petals has a free tier too, but ChatGPT's free offer is broadly known and trusted.

Where Petals is different

The central question is not recall rate — it is what you can do with your memory after the AI has built it.

ChatGPT's memory lives in a list and a synthesis document you read but cannot fully verify. Petals' memory lives in a graph you can navigate. Open any node: see what the system believes, where it came from, what it connects to. Disagree with something — edit it. A fact has changed — correct it. A relationship is wrong — delete it. None of this requires trusting that the background process got it right.

The second difference is action. Petals pairs memory with connectors, automations, and parallel agents. Your assistant can reach into your apps, run scheduled tasks, and propose automations for your approval. ChatGPT has Tasks, but it does not pair deep memory with a structured permission model for taking actions on your behalf.

The third difference is price. ChatGPT Plus is $20/month. Petals Plus is $9/month. Both include memory. The use cases are different enough that this is not a straight substitution — but if Petals fits your workflow, it costs less.

The short version

If you want image generation, real-time voice, and a memory that maintains itself automatically, ChatGPT is worth what it charges.

If you want to see the shape of what your assistant knows — browse it, correct it, build on it — Petals is the one that shows you. It remembers so you can think.

Start free — or read about how it's priced.