CrawlSpider

The AI knowledge reference

LLM knowledge cutoff dates.
Know where the knowledge ends.

Compare knowledge cutoff dates across 16 AI providers, including GPT, Claude, Gemini, Grok, Llama, DeepSeek, Qwen and Kimi. See when models can look beyond training with web search.

Explore the model directory What is a knowledge cutoff?
Source review: Official provider sources
50model entries
16AI providers
4assistant references

Compare the models

The knowledge cutoff directory

Click a column heading to sort. Knowledge age measures the gap from cutoff to release; month-only cutoffs are approximate.

Filter by provider

Showing 50 of 50 models

Reset filters

Swipe across the table to compare all columns

LLM knowledge cutoff dates, web access, releases, lifecycle status, official sources and last verification dates.
Provider / model Knowledge cutoff Search / web accessRelease date Knowledge age at release StatusSources / last verified
CohereCommand A+command-a-plus-05-2026
 Apr 1, 2025Knowledge cutoff
Context

The linked model documentation states this cutoff.

 Retrieval / toolsTool use and supplied retrieval can add fresh information; no always-on web search is implied. May 20, 2026Official release date 1 year, 1 monthCutoff to release Active
Details

Listed in the reviewed official documentation; availability can depend on deployment.
1 official source
GoogleGemini 3.1 Pro Previewgemini-3.1-pro-preview
 Jan 2025Family-level cutoff
Context

January 2025 in the Gemini 3 developer guide; a family-level statement.

 Google Search groundingOptional; must be enabled. Feb 19, 2026Preview release Approx. 1 year, 1 monthCutoff to release Preview
Details

No shutdown date announced in the reviewed schedule.
3 official sources
MetaLlama 3.3 70B Instructllama-3.3-70b-instruct
 Dec 2023Knowledge cutoff
Context

The official documentation states this cutoff.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Dec 6, 2024Official model release Approx. 1 yearCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
GoogleGemini 3 Flash Previewgemini-3-flash-preview
 Jan 2025Family-level cutoff
Context

January 2025 in the Gemini 3 developer guide; a family-level statement.

 Google Search groundingOptional; must be enabled. Dec 17, 2025Preview release Approx. 11 monthsCutoff to release Preview
Details

No shutdown date announced in the reviewed schedule.
3 official sources
NVIDIANemotron 3.5 Lightning 30B A3Bnemotron-3.5-lightning-30b-a3b
 Sep 2025Pre-training data cutoff
Context

NVIDIA states a pre-training cutoff of 2025-09 and a post-training cutoff of 2026-05. The table uses pre-training for the age comparison. Post-training freshness does not establish a universal knowledge cutoff.

 Application-dependentListed in NVIDIA NIM. Fresh web information requires search or retrieval supplied by the application; hosting does not refresh training. Aug 11, 2026Official model release Approx. 11 monthsCutoff to release Open weightsNIM listed
Details

NVIDIA-developed open-weight model listed in the NIM catalogue. License and deployment requirements are in the linked model card.
2 official sources
NVIDIANemotron 3 Super 120B A12Bnemotron-3-super-120b-a12b
 Jun 2025Pre-training data cutoff
Context

NVIDIA states a pre-training cutoff of 2025-06 and a post-training cutoff of 2026-02. The table uses pre-training for the age comparison. Post-training freshness does not establish a universal knowledge cutoff.

 Application-dependentListed in NVIDIA NIM. Fresh web information requires search or retrieval supplied by the application; hosting does not refresh training. Mar 11, 2026Official model release Approx. 9 monthsCutoff to release Open weightsNIM listed
Details

NVIDIA-developed open-weight model listed in the NIM catalogue. License and deployment requirements are in the linked model card.
2 official sources
NVIDIANemotron 3 Ultra 550B A55Bnemotron-3-ultra-550b-a55b
 Sep 2025Pre-training data cutoff
Context

NVIDIA states a pre-training cutoff of 2025-09 and a post-training cutoff of 2026-05. The table uses pre-training for the age comparison. Post-training freshness does not establish a universal knowledge cutoff.

 Application-dependentListed in NVIDIA NIM. Fresh web information requires search or retrieval supplied by the application; hosting does not refresh training. Jun 4, 2026Official model release Approx. 9 monthsCutoff to release Open weightsNIM listed
Details

NVIDIA-developed open-weight model listed in the NIM catalogue. License and deployment requirements are in the linked model card.
2 official sources
AnthropicClaude Haiku 4.5claude-haiku-4-5-20251001
 Feb 2025Reliable knowledge cutoff
Context

Training data extends to July 2025; reliable knowledge has an earlier cutoff. Training-data cutoff: Jul 2025.

 Optional web searchAvailability depends on platform and settings. Oct 15, 2025Provider release Approx. 8 monthsCutoff to release Active
Details

Provider lists this model as active.
3 official sources
MetaLlama 4 Maverick (17Bx128E)llama-4-maverick
 Aug 2024Knowledge cutoff
Context

The linked model documentation states this cutoff.

 Application-dependentStatic offline model. A host application can supply newer information. Apr 5, 2025Official release date Approx. 8 monthsCutoff to release Open weights
Details

Published weights under the Llama 4 Community License; this is not a claim about a hosted API endpoint.
1 official source
MetaLlama 4 Scout (17Bx16E)llama-4-scout
 Aug 2024Knowledge cutoff
Context

The linked model documentation states this cutoff.

 Application-dependentStatic offline model. A host application can supply newer information. Apr 5, 2025Official release date Approx. 8 monthsCutoff to release Open weights
Details

Published weights under the Llama 4 Community License; this is not a claim about a hosted API endpoint.
1 official source
GoogleGemini 2.5 Flash-Litegemini-2.5-flash-lite
 Jan 2025Knowledge cutoff
Context

Provider-published date.

 Google Search groundingOptional; must be enabled. Jul 22, 2025Stable API release Approx. 6 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
2 official sources
GoogleGemini 3.8 Flashgemini-3.8-flash
 Mar 2026Knowledge cutoff
Context

Some subject areas remain limited to January 2025.

 Google Search groundingOptional; must be enabled. Sep 2, 2026Stable API release Approx. 6 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
3 official sources
AnthropicClaude Sonnet 5claude-sonnet-5
 Jan 2026Reliable knowledge cutoff
Context

Reliable knowledge and training-data cutoffs match. Training-data cutoff: Jan 2026.

 Optional web searchAvailability depends on platform and settings. Jun 30, 2026Provider release Approx. 5 monthsCutoff to release Active
Details

Provider lists this model as active.
3 official sources
NVIDIANemotron 3 Nano 30B A3Bnemotron-3-nano-30b-a3b
 Jun 25, 2025Pre-training data cutoff
Context

Pre-training cutoff: June 25, 2025. Post-training cutoff: November 28, 2025. The table uses pre-training for its age comparison.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Dec 15, 2025Official model release 5 monthsCutoff to release Open weightsNIM listed
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
2 official sources
GoogleGemini 2.5 Flashgemini-2.5-flash
 Jan 2025Knowledge cutoff
Context

Provider-published date.

 Google Search groundingOptional; must be enabled. Jun 17, 2025Stable API release Approx. 5 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
2 official sources
GoogleGemini 2.5 Progemini-2.5-pro
 Jan 2025Knowledge cutoff
Context

Provider-published date.

 Google Search groundingOptional; must be enabled. Jun 17, 2025Stable API release Approx. 5 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
2 official sources
GoogleGemini 3.7 Flashgemini-3.7-flash
 Mar 2026Knowledge cutoff
Context

Some subject areas remain limited to January 2025.

 Google Search groundingOptional; must be enabled. Aug 13, 2026Stable API release Approx. 5 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
3 official sources
OpenAIGPT-5.6 Lunagpt-5.6-luna
 Feb 16, 2026Knowledge cutoff
Context

Provider-published date.

 Optional web searchRequires a web search tool. Jul 9, 2026API release 4 monthsCutoff to release Active
Details

Listed in the current catalog.
3 official sources
OpenAIGPT-5.6 Solgpt-5.6-sol
 Feb 16, 2026Knowledge cutoff
Context

Provider-published date.

 Optional web searchRequires a web search tool. Jul 9, 2026API release 4 monthsCutoff to release Active
Details

Also available as gpt-5.6.
3 official sources
OpenAIGPT-5.6 Terragpt-5.6-terra
 Feb 16, 2026Knowledge cutoff
Context

Provider-published date.

 Optional web searchRequires a web search tool. Jul 9, 2026API release 4 monthsCutoff to release Active
Details

Listed in the current catalog.
3 official sources
GoogleGemini 3.5 Flash-Litegemini-3.5-flash-lite
 Mar 2026Knowledge cutoff
Context

Some subject areas remain limited to January 2025.

 Google Search groundingOptional; must be enabled. Jul 21, 2026Stable API release Approx. 4 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
3 official sources
GoogleGemini 3.6 Flashgemini-3.6-flash
 Mar 2026Knowledge cutoff
Context

Some subject areas remain limited to January 2025.

 Google Search groundingOptional; must be enabled. Jul 21, 2026Stable API release Approx. 4 monthsCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
3 official sources
OpenAIGPT-6 Astragpt-6-astra
 Apr 30, 2026Knowledge cutoff
Context

Provider-published date.

 Optional web searchRequires a web search tool. Sep 3, 2026API release 4 monthsCutoff to release Active
Details

Listed in the current catalog.
3 official sources
AnthropicClaude Fable 5.1claude-fable-5-1
 Jun 2026Reliable knowledge cutoff
Context

Reliable knowledge and training-data cutoffs match. Training-data cutoff: Jun 2026.

 Optional web searchAvailability depends on platform and settings. Sep 1, 2026Provider release Approx. 3 monthsCutoff to release Active
Details

Provider lists this model as active.
3 official sources
AnthropicClaude Opus 5claude-opus-5
 May 2026Reliable knowledge cutoff
Context

Reliable knowledge and training-data cutoffs match. Training-data cutoff: May 2026.

 Optional web searchAvailability depends on platform and settings. Jul 24, 2026Provider release Approx. 2 monthsCutoff to release Active
Details

Provider lists this model as active.
3 official sources
DeepSeekDeepSeek R1deepseek-r1
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weightsNIM listed
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
2 official sources
DeepSeekDeepSeek V4 Prodeepseek-v4-pro
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Not statedSee source notes Not statedCutoff to release Active
Details

The current quick-start explicitly says V4 Pro service continues after September 14, 2026. Older migration notes can differ; check the current provider notice.
1 official source
DeepSeekDeepSeek V4.1 Flashdeepseek-flash
 Not statedNot disclosed
Context

No cutoff stated in the reviewed API guide. The deepseek-flash alias currently serves V4.1 Flash; legacy V4 Flash aliases also route here.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Not statedSee source notes Not statedCutoff to release Active
Details

Listed in the reviewed official documentation; availability can depend on deployment.
1 official source
Z.aiGLM-5glm-5
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
GoogleGemini 3.1 Flash-Litegemini-3.1-flash-lite
 Not statedKnowledge cutoff
Context

No model-specific cutoff stated in the documentation and model card reviewed.

 Google Search groundingOptional; must be enabled. May 7, 2026Stable API release Not statedCutoff to release Stable
Details

Earliest listed shutdown: May 7, 2027.
3 official sources
GoogleGemini 3.5 Flashgemini-3.5-flash
 Not statedKnowledge cutoff
Context

No model-specific cutoff stated in the documentation and model card reviewed.

 Google Search groundingOptional; must be enabled. May 19, 2026Stable API release Not statedCutoff to release Stable
Details

No shutdown date announced in the reviewed schedule.
3 official sources
GoogleGemma 3 27B Instructgemma-3-27b-it
 Aug 2024Knowledge cutoff
Context

The official documentation states this cutoff.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weightsNIM listed
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
2 official sources
IBMGranite 3.3 8B Instructgranite-3.3-8b-instruct
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
xAIGrok 4.6grok-4.6
 Feb 1, 2026Knowledge cutoff
Context

The linked model documentation states this cutoff.

 Optional searchWeb Search and X Search must be enabled for real-time information. Not statedSee source notes Not statedCutoff to release Active
Details

Listed in the reviewed official documentation; availability can depend on deployment.
1 official source
Moonshot AIKimi K2.6kimi-k2.6
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Not statedSee source notes Not statedCutoff to release Active
Details

Listed in the reviewed official documentation; availability can depend on deployment.
1 official source
Moonshot AIKimi K3kimi-k3
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Not statedSee source notes Not statedCutoff to release Active
Details

Listed in the reviewed official documentation; availability can depend on deployment.
1 official source
MiniMaxMiniMax M2.5minimax-m2.5
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
MistralMistral Medium 3.5mistral-medium-3-5
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Apr 28, 2026Official release date Not statedCutoff to release Active
Details

Generally available; also released as open weights under Modified MIT.
1 official source
MistralMistral Small 4mistral-small-4-119b-2603
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
MetaMuse Spark 1.3muse-spark-1.3
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Tool-dependentFresh web information requires tools or retrieval configured by the application. Sep 2, 2026Official release date Not statedCutoff to release Active
Details

Released in Muse Code and Meta Model API. The announcement describes open weights as forthcoming.
1 official source
MicrosoftPhi-4 Mini Instructphi-4-mini-instruct
 Jun 2024Knowledge cutoff
Context

June 2024 is the stated cutoff for publicly available training data. The card gives a February 2025 release month; an exact release day is not established here.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weightsNIM listed
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
2 official sources
Alibaba / QwenQwen3 Coder 480B A35B Instructqwen3-coder-480b-a35b-instruct
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed model card. Release and upload dates are not treated as cutoffs.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source
Alibaba / QwenQwen3.8-Flash-Nextqwen3.8-flash-next
 Not statedNot disclosed
Context

No model-specific knowledge cutoff is stated in the reviewed source. Release dates and upload timestamps are not training cutoffs.

 Application-dependentDownloaded weights need external tools for web retrieval. The hosted Qwen3.8-Flash service is a separate offering. Not statedSee source notes Not statedCutoff to release Open weights
Details

Open-weight experimental architecture preview under Qwen Community 1.0.
1 official source
PerplexitySonarsonar
 Not statedNot disclosed
Context

No training cutoff stated in the reviewed Sonar documentation.

 Search-groundedRetrieves web sources for answers. Not statedSee source notes Not statedCutoff to release Legacy API
Details

Listed under the legacy Sonar API; this is not a retirement notice.
3 official sources
PerplexitySonar Deep Researchsonar-deep-research
 Not statedNot disclosed
Context

No training cutoff stated in the reviewed Sonar documentation.

 Search-groundedRetrieves web sources for answers. Not statedSee source notes Not statedCutoff to release Legacy API
Details

Listed under the legacy Sonar API; this is not a retirement notice.
3 official sources
PerplexitySonar Prosonar-pro
 Not statedNot disclosed
Context

No training cutoff stated in the reviewed Sonar documentation.

 Search-groundedRetrieves web sources for answers. Not statedSee source notes Not statedCutoff to release Legacy API
Details

Listed under the legacy Sonar API; this is not a retirement notice.
3 official sources
PerplexitySonar Reasoningsonar-reasoning
 Not statedNot disclosed
Context

No verified training cutoff recorded.

 Historical entryUse Sonar Reasoning Pro instead. Not statedSee source notes Not statedCutoff to release Deprecated
Details

Deprecated December 15, 2025.
1 official source
PerplexitySonar Reasoning Prosonar-reasoning-pro
 Not statedNot disclosed
Context

No training cutoff stated in the reviewed Sonar documentation.

 Search-groundedRetrieves web sources for answers. Not statedSee source notes Not statedCutoff to release Legacy API
Details

Listed under the legacy Sonar API; this is not a retirement notice.
3 official sources
OpenAIgpt-oss-120bgpt-oss-120b
 Jun 1, 2024Knowledge cutoff
Context

The official documentation states this cutoff.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weightsNIM listed
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
2 official sources
OpenAIgpt-oss-20bgpt-oss-20b
 Jun 1, 2024Knowledge cutoff
Context

The official documentation states this cutoff.

 Application-dependentNewer web information requires search or retrieval tools supplied by the host application. Not statedSee source notes Not statedCutoff to release Open weights
Details

Published open weights; check the official model card for licensing. NIM listing is tracked separately from model authorship.
1 official source

Coverage: Selected text, chat, coding and reasoning models from 16 developers, including open-weight families GPT-OSS, Gemma, Phi, Granite, Nemotron, GLM and MiniMax. This is not the full NIM inventory. NIM listing is verified separately; unverified listings remain unknown. Specialized embedding, reranking, reward and safety models are excluded. Assistant products are compared separately. Dates describe the listed model or API endpoint, not every answer in a consumer app.

Models power products

Copilot, Meta AI & other AI assistants

An assistant can combine models, search and connected information. Its name alone does not tell you a training cutoff.

Microsoft

Microsoft Copilot

No single product-wide cutoff. The system can use models from different developers; check the selected model and mode.

Fresh information

Can ground answers in web search results and provide source links.

Microsoft Copilot transparency note ↗ (opens in a new tab)
Microsoft

Microsoft 365 Copilot Chat

A product and retrieval experience, not one fixed model with a universal cutoff.

Fresh information

Bing search is available when enabled. Organization policies and user settings control access.

Microsoft 365 Copilot web search ↗ (opens in a new tab)
Meta

Meta AI

Do not apply Llama 4’s August 2024 cutoff to every Meta AI answer. Meta publishes multiple model families; the reviewed sources do not establish one assistant-wide cutoff.

Fresh information

Check the sources and mode in the actual assistant. A Muse API release does not establish the model used for every Meta AI conversation.

Meta model catalogue ↗ (opens in a new tab)Meta Llama 4 model card ↗ (opens in a new tab)Meta Muse Spark 1.3 release ↗ (opens in a new tab)
Mistral

Vibe (formerly Le Chat)

Le Chat became Vibe on August 12, 2026. Its Work, Code and Chat modes are product experiences, not a single model cutoff.

Fresh information

Connected apps and tools can supply context. Check the mode and sources used for each answer.

Le Chat is now Vibe ↗ (opens in a new tab)

A quick guide

Three dates. Three different meanings.

Use the right date to answer the right question.

01

Knowledge cutoff

The provider's stated boundary for built-in knowledge. It does not mean a model knows every fact published before that date.

“What might it know from training?”
02

Release date

When the listed model became available. A new release can still have an older cutoff. Stable and preview releases are labeled separately.

“When could I start using it?”
03

Last verified

When CrawlSpider reviewed the linked documentation. This date records our source check; it is not a model update or a fresh training run.

“When was this entry checked?”

From knowledge to visibility

An older cutoff doesn't have to leave your brand behind.

Search-enabled models can use newer pages when answering a buyer's question. Your brand's presence also depends on what gets retrieved and how the answer uses it.

Use cutoff dates as context for an LLM visibility audit. Then check the questions your customers ask, the competitors that appear, and how your brand is described.

Check my brand's AI visibility
What can inform an answer?
Built-in knowledgeBounded by the model's training
Retrieved web contentWhen search is enabled and used

The answer your buyer sees

Sources & methodology

A reference you can verify.

We use provider documentation, model cards, and release notes. Every row links to its evidence. Missing dates remain “Not stated”; search access is recorded separately.

For Claude, the main date is the reliable-knowledge cutoff, with training-data dates under Context. For Gemini, model-specific dates take priority over broad family statements. For NVIDIA Nemotron, the table uses the stated pre-training data cutoff; post-training dates appear under Context. NIM is a serving platform and also offers other providers’ models. A month stays a month—we do not invent an exact day.

OpenAI docs (opens in a new tab)Anthropic docs (opens in a new tab)Google docs (opens in a new tab)Perplexity docs (opens in a new tab)xAI docs (opens in a new tab)Meta docs (opens in a new tab)DeepSeek docs (opens in a new tab)Mistral docs (opens in a new tab)Alibaba / Qwen docs (opens in a new tab)Moonshot AI docs (opens in a new tab)Cohere docs (opens in a new tab)NVIDIA docs (opens in a new tab)Microsoft docs (opens in a new tab)IBM docs (opens in a new tab)Z.ai docs (opens in a new tab)MiniMax docs (opens in a new tab)

Common questions

Making sense of model cutoffs

What is ChatGPT's knowledge cutoff date?

There is no single date for every model used in ChatGPT. Check the exact model and version, and whether search was used for the answer. The OpenAI rows above describe the listed API models; they do not establish which model served a particular ChatGPT conversation.

Can an AI model know about events after its cutoff?

It can use newer information supplied in your prompt, an uploaded document, or retrieved web content. For example, Claude's web search documentation explains how search can provide information beyond the model's cutoff. Access depends on the product, settings, and tools available.

Why doesn't Perplexity have a simple cutoff date?

Sonar combines answer generation with web retrieval. The Sonar documentation describes search-grounded answers, but the sources reviewed for this table do not state a training cutoff. “Search-grounded” describes how an answer gets information, not when the model finished training.

What is Grok's knowledge cutoff?

The reviewed xAI documentation lists February 1, 2026 for Grok 4.6. Web Search and X Search can supply newer information when enabled. Check the Grok row for its official source and verification date.

Do Llama and Meta AI have the same cutoff?

Not necessarily. Meta's Llama 4 model card states August 2024 for Scout and Maverick. Meta AI is an assistant product, and Meta also publishes Muse models. A Llama model's cutoff should not be assigned to the whole assistant.

What is Microsoft Copilot's knowledge cutoff?

There is no single cutoff established for all Copilot experiences. Identify the product, selected model and whether web search was used. The assistant comparison above links to Microsoft's explanation of model selection and web grounding.

Does this list every provider and every model?

No. This edition covers selected models from 16 providers, plus separate assistant references. It excludes most historical snapshots, specialized media models and private systems. Use the official catalogues to explore models outside this page's scope.

Does a newer cutoff mean better AI visibility for my brand?

A cutoff alone cannot tell you whether a model will mention your business. The question, model, source retrieval, and information about your brand all affect the answer. Run a visibility check and compare repeated results to understand where your brand appears.

Directory updates

Added 12 selected open-weight models across existing and four new developers. Added a filter for verified NIM listings; unverified hosting is not treated as unavailable.

Added three NVIDIA Nemotron open-weight models listed in NIM. Pre-training and post-training data dates are distinguished; previous entries retain their verification dates.

Expanded to 11 providers and 35 model entries. Added Copilot, Microsoft 365 Copilot Chat, Meta AI and Vibe (formerly Le Chat) as separate assistant references.

First edition: 24 model entries across four providers, with official sources, cutoff qualifications, and lifecycle notes.

See the bigger picture

Does AI know your brand?
Find out where you stand.

Start with a free visibility snapshot and see which brands appear in answers to buyer questions.

Get my free visibility check No signup to start. No credit card required.