- Home
- What is a training cutoff?
What is a training cutoff?
Short answer: an AI model like ChatGPT, Claude or Gemini learns from a huge snapshot of text collected up to a certain date, its training cutoff (also called knowledge cutoff). It knows nothing that happened after that date unless someone tells it, for example through web search. Models are released months after their cutoff and used for a year or more, so the gap between what a model knows and what is true today keeps growing.
Where the cutoff comes from
A language model is built in stages. First, companies collect training data: mostly web pages (often starting from the Common Crawl archive[5]), plus books, code and licensed or purchased data. GPT-3's paper, for example, describes filtered Common Crawl data from 2016 to 2019[4]. At some point the collection stops. That moment is the training cutoff.
Then the model is trained on that frozen snapshot, which takes weeks to months, and afterwards tuned, safety-tested and packaged into products. So a model is always released months after its cutoff, and then used for a year or more. Anthropic, for example, lists retirement dates at least a year after release[1]. Nothing the model learned is updated in the meantime: its knowledge is fixed in its weights.
The gap for today's models
From our model registry: cutoff, release date and the gap between them (then add however long you've been using it).
| Model | Made by | Knows news until | Released | Gap |
|---|---|---|---|---|
| GPT-6.1 Sol | OpenAI | Apr 2026 | 2026-09-29 | 5 mo |
| Claude Sonnet 5.5 | Anthropic | Jun 2026 | 2026-09-28 | 3 mo |
| Claude Opus 5.5 | Anthropic | Jun 2026 | 2026-09-22 | 3 mo |
| GPT-6 Luna | OpenAI | May 2026 | 2026-09-22 | 4 mo |
| GPT-6 Sol | OpenAI | Apr 2026 | 2026-09-22 | 5 mo |
| Grok 4.7 | xAI | May 2026 | 2026-09-21 | 4 mo |
| GPT-6 Astra | OpenAI | Apr 2026 | 2026-09-03 | 5 mo |
| Gemini 3.8 Flash | Google DeepMind | Mar 2026 | 2026-09-02 | 6 mo |
| Claude Fable 5.1 | Anthropic | Jun 2026 | 2026-09-01 | 3 mo |
| Claude Mythos 5.1 | Anthropic | Jun 2026 | 2026-09-01 | 3 mo |
| Muse Glimmer 30B | Meta | Jan 2026 | 2026-08 | 7 mo |
| GPT-5.6 Terra | OpenAI | Feb 2026 | 2026-07-09 | 5 mo |
| Claude Haiku 4.5 | Anthropic | Feb 2025 | 2025-10-15 | 8 mo |
| gpt-oss-120b | OpenAI | Jun 2024 | 2025-08-05 | 14 mo |
| gpt-oss-20b | OpenAI | Jun 2024 | 2025-08-05 | 14 mo |
Cutoffs as published by each lab[1][2]. Full list by model: briefings by model.
The cutoff is blurrier than one date
The internet keeps writing about an event for months and years after it happens: analysis, follow-ups, Wikipedia edits. So at training time the last few months before the cutoff are only thinly covered, and the model knows them less reliably than older events. Anthropic publishes two dates for this reason, a "training data cutoff" and an earlier "reliable knowledge cutoff". For Claude Haiku 4.5 they are July 2025 and February 2025[1].
Researchers have also found that a model's effective cutoff often differs from the reported one, partly because web crawls contain older copies of pages and because of how data is deduplicated[7]. And models get measurably worse at text from after their training period, more so the further it lies in the future[6]. Time even leaves a trace inside the model's weights[11].
Why models don't know today's date
A model has no clock. On its own it tends to assume that "now" is roughly its training cutoff. Chat apps work around this by quietly giving the model the current date in a hidden instruction (the system prompt) at the start of every conversation[3]. Through the raw API, or in tools that don't do this, the model may not know what day it is at all.
Why models call real news fake
Knowing the date isn't enough. When a model meets news from after its cutoff, such as a new pope, a new president, or a newer AI model with an unfamiliar name, it often concludes the news must be invented, rather than that time has passed. Real examples:
- 2023: Bing Chat insisted it was still 2022 and that a film already in cinemas wasn't out yet. Case
- 2025: several chatbots said "Pope Leo XIV does not exist" months after his election. Case
- 2025: an early build of Gemini 3 refused to believe it was 2025 and called real news articles AI-generated. Case
- 2026: we gave Gemini 3.1 Pro our own sourced timeline. Offline it called it fiction in 9 of 10 runs, even when told the date; with Google Search on it said "I'm stunned!" and accepted it. Case
We call this cutoff blindness. All 22 documented cases, with sources: Why this exists.
How to fix it
- Turn on web search when your chat app offers it. Letting a model look things up (retrieval) is the standard way to give it fresh facts[9], and search results measurably improve answers about current events[8][10]. In our test, search was the only thing that reliably worked.
- Tell it today's date if it might not know (most apps already do).
- Give it a briefing of what it missed. That's what this site is for: paste postcutoff.com/ai into the chat, or pick the file for your model on briefings by model.
Sources
- Anthropic, Models overview: "Reliable knowledge cutoff" and "Training data cutoff" per model
- OpenAI, Models: each model page lists its knowledge cutoff
- Anthropic, System prompts: the apps give Claude "up-to-date information, such as the current date"
- Brown et al. (2020), Language Models are Few-Shot Learners (GPT-3; training data from Common Crawl 2016–2019)
- Common Crawl, the open web archive most training sets start from
- Lazaridou et al. (2021), Mind the Gap: Assessing Temporal Generalization in Neural Language Models
- Cheng, Marone, Weller et al. (2024), Dated Data: Tracing Knowledge Cutoffs in Large Language Models
- Kasai et al. (2022), RealTime QA: What's the Answer Right Now?
- Lewis et al. (2020), Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
- Vu et al. (2023), FreshLLMs: Refreshing Large Language Models with Search Engine Augmentation
- Nylund, Gururangan, Smith (2023), Time is Encoded in the Weights of Finetuned Language Models