Knowledge cutoffs - verified
AI model knowledge cutoff dates
66 of the 167 models in the registry have a knowledge cutoff stated by their provider, across 6 providers. Seed 2.0 Pro and MiniMax-M3 publish only a training-data date, listed as "Training data through". The other 99 are listed as not stated, because their providers' pages give no date.
Every date links to the provider's own page. What the dates mean - Models with no stated cutoff
- Stated cutoffs
- 66
- Providers
- 6
- Training data only
- 2
- Not stated
- 99
Every stated cutoff
Newest cutoff first, in the provider's own precision. The gap is whole calendar months from the cutoff to the model's release. Training data cutoff is a separate date only some providers publish.
68 models
| Model | Provider | Knowledge cutoff | Training data cutoff | Released | Gapcutoff to release | Source |
|---|---|---|---|---|---|---|
| No models match these filters. | ||||||
Claude Haiku 5.5claude-haiku-5-5 | Anthropic | Reliable knowledge cutoff | 4 mo | platform.claude.com | ||
Claude Sonnet 5.5claude-sonnet-5-5 | Anthropic | Reliable knowledge cutoff | 3 mo | platform.claude.com | ||
Claude Opus 5.5claude-opus-5-5 | Anthropic | Reliable knowledge cutoff | 3 mo | platform.claude.com | ||
Claude Fable 5.1claude-fable-5-1 | Anthropic | Reliable knowledge cutoff | 3 mo | platform.claude.com | ||
Claude Mythos 5.1claude-mythos-5-1 | Anthropic | Reliable knowledge cutoff | 3 mo | platform.claude.com | ||
GPT-6 Lunagpt-6-luna | OpenAI | -No separate training data date stated | 4 mo | developers.openai.com | ||
Grok 4.7grok-4.7 | xAI | -No separate training data date stated | 4 mo | docs.x.ai | ||
Claude Opus 5claude-opus-5 | Anthropic | Reliable knowledge cutoff | 2 mo | platform.claude.com | ||
GPT-6.1 Solgpt-6.1-sol | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
GPT-6 Astragpt-6-astra | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
GPT-6 Solgpt-6-sol | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
Chat Latestchat-latestNote: chat-latest is an alias that OpenAI repoints to the current ChatGPT Instant snapshot, so this date moves when the alias does. | OpenAI | -No separate training data date stated | Not stated | -Not computed | developers.openai.com | |
Gemini 4 Argongemini-4-argonNote: The Gemini 4 Argon model card states this date for Gemini 4 as a whole; the Gemini API model page lists no cutoff. | -No separate training data date stated | 7 mo | deepmind.google | |||
Gemini 3.8 Flashgemini-3.8-flashNote: Google says knowledge in some domains may be limited to January 2025, in line with the Gemini 3 family. Source is the Google DeepMind model card. | -No separate training data date stated | 6 mo | deepmind.google | |||
Gemini 3.5 Flash-Litegemini-3.5-flash-liteNote: Google says knowledge in some domains may be limited to January 2025, in line with the Gemini 3 family. Source is the Google DeepMind model card. | -No separate training data date stated | 4 mo | deepmind.google | |||
Gemini 3.6 Flashgemini-3.6-flashNote: Google says knowledge in some domains may be limited to January 2025, in line with the Gemini 3 family. Source is the Google DeepMind model card. | -No separate training data date stated | 4 mo | deepmind.google | |||
GPT-5.6 Cybergpt-5.6-cyber | OpenAI | -No separate training data date stated | 6 mo | developers.openai.com | ||
GPT-5.6 Lunagpt-5.6-luna | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
GPT-5.6 Solgpt-5.6-sol | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
GPT-5.6 Terragpt-5.6-terra | OpenAI | -No separate training data date stated | 5 mo | developers.openai.com | ||
Claude Sonnet 5claude-sonnet-5 | Anthropic | Reliable knowledge cutoff | 5 mo | platform.claude.com | ||
Claude Fable 5claude-fable-5 | Anthropic | Reliable knowledge cutoff | 5 mo | platform.claude.com | ||
Claude Mythos 5claude-mythos-5 | Anthropic | Reliable knowledge cutoff | 5 mo | platform.claude.com | ||
Claude Opus 4.8claude-opus-4-8 | Anthropic | Reliable knowledge cutoff | 4 mo | platform.claude.com | ||
Claude Opus 4.7claude-opus-4-7 | Anthropic | Reliable knowledge cutoff | 3 mo | platform.claude.com | ||
GPT-5.5gpt-5.5 | OpenAI | -No separate training data date stated | 4 mo | developers.openai.com | ||
GPT-5.5 Progpt-5.5-pro | OpenAI | -No separate training data date stated | 4 mo | developers.openai.com | ||
Amazon Nova 2 Liteamazon.nova-2-lite-v1:0 | Amazon | -No separate training data date stated | 2 mo | docs.aws.amazon.com | ||
GPT-5.4 Minigpt-5.4-mini | OpenAI | -No separate training data date stated | 7 mo | developers.openai.com | ||
GPT-5.4 nanogpt-5.4-nano | OpenAI | -No separate training data date stated | 7 mo | developers.openai.com | ||
GPT-5.4gpt-5.4 | OpenAI | -No separate training data date stated | 7 mo | developers.openai.com | ||
GPT-5.3-Codexgpt-5.3-codex | OpenAI | -No separate training data date stated | 6 mo | developers.openai.com | ||
GPT-5.2gpt-5.2 | OpenAI | -No separate training data date stated | 4 mo | developers.openai.com | ||
Claude Sonnet 4.6claude-sonnet-4-6 | Anthropic | Reliable knowledge cutoff | 6 mo | platform.claude.com | ||
Claude Opus 4.6claude-opus-4-6 | Anthropic | Reliable knowledge cutoff | 9 mo | platform.claude.com | ||
Claude Opus 4.5claude-opus-4-5-20251101 | Anthropic | Reliable knowledge cutoff | 6 mo | platform.claude.com | ||
Claude Haiku 4.5claude-haiku-4-5-20251001 | Anthropic | Reliable knowledge cutoff | 8 mo | platform.claude.com | ||
Claude Sonnet 4.5claude-sonnet-4-5-20250929 | Anthropic | Reliable knowledge cutoff | 8 mo | platform.claude.com | ||
Gemini 2.5 Flash-Litegemini-2.5-flash-lite | -No separate training data date stated | 6 mo | ai.google.dev | |||
Gemini 2.5 Flashgemini-2.5-flash | -No separate training data date stated | 5 mo | ai.google.dev | |||
Gemini 2.5 Progemini-2.5-pro | -No separate training data date stated | 5 mo | ai.google.dev | |||
Amazon Nova Liteamazon.nova-lite-v1:0Note: AWS's AI Service Card for Nova separately says it does not define a cutoff date for training; this is the Bedrock model card value. | Amazon | -No separate training data date stated | 2 mo | docs.aws.amazon.com | ||
Amazon Nova Microamazon.nova-micro-v1:0Note: AWS's AI Service Card for Nova separately says it does not define a cutoff date for training; this is the Bedrock model card value. | Amazon | -No separate training data date stated | 2 mo | docs.aws.amazon.com | ||
Amazon Nova Proamazon.nova-pro-v1:0Note: AWS's AI Service Card for Nova separately says it does not define a cutoff date for training; this is the Bedrock model card value. | Amazon | -No separate training data date stated | 2 mo | docs.aws.amazon.com | ||
GPT-5.1gpt-5.1 | OpenAI | -No separate training data date stated | 14 mo | developers.openai.com | ||
GPT-5 Progpt-5-pro | OpenAI | -No separate training data date stated | 13 mo | developers.openai.com | ||
GPT-5gpt-5 | OpenAI | -No separate training data date stated | 11 mo | developers.openai.com | ||
o3-proo3-pro | OpenAI | -No separate training data date stated | 12 mo | developers.openai.com | ||
o3o3 | OpenAI | -No separate training data date stated | 10 mo | developers.openai.com | ||
o4-minio4-mini | OpenAI | -No separate training data date stated | 10 mo | developers.openai.com | ||
GPT-4.1gpt-4.1 | OpenAI | -No separate training data date stated | 10 mo | developers.openai.com | ||
GPT-4.1 Minigpt-4.1-mini | OpenAI | -No separate training data date stated | 10 mo | developers.openai.com | ||
GPT-4.1 nanogpt-4.1-nano | OpenAI | -No separate training data date stated | 10 mo | developers.openai.com | ||
Command Acommand-a-03-2025Note: Cohere lists the same June 1, 2024 cutoff on every Command model page. | Cohere | -No separate training data date stated | 9 mo | docs.cohere.com | ||
Command R7Bcommand-r7b-12-2024Note: Cohere lists the same June 1, 2024 cutoff on every Command model page. | Cohere | -No separate training data date stated | 6 mo | docs.cohere.com | ||
GPT-5 Minigpt-5-mini | OpenAI | -No separate training data date stated | 15 mo | developers.openai.com | ||
GPT-5 nanogpt-5-nano | OpenAI | -No separate training data date stated | 15 mo | developers.openai.com | ||
GPT-4 Turbogpt-4-turbo | OpenAI | -No separate training data date stated | 4 mo | developers.openai.com | ||
GPT-4gpt-4 | OpenAI | -No separate training data date stated | Not stated | -Not computed | developers.openai.com | |
o1-proo1-pro | OpenAI | -No separate training data date stated | 17 mo | developers.openai.com | ||
o3-minio3-mini | OpenAI | -No separate training data date stated | 15 mo | developers.openai.com | ||
o1o1 | OpenAI | -No separate training data date stated | 14 mo | developers.openai.com | ||
GPT-4o Minigpt-4o-mini | OpenAI | -No separate training data date stated | 9 mo | developers.openai.com | ||
GPT-4ogpt-4o | OpenAI | -No separate training data date stated | 7 mo | developers.openai.com | ||
GPT-3.5 Turbogpt-3.5-turbo | OpenAI | -No separate training data date stated | Not stated | -Not computed | developers.openai.com | |
GPT-3.5 Turbo Instructgpt-3.5-turbo-instruct | OpenAI | -No separate training data date stated | Not stated | -Not computed | developers.openai.com | |
Seed 2.0 Proseed-2-0-pro-260328Note: EU AI Act Public Summary for 'Seed 2.0-Pro (Model name on Byteplus: Dola-Seed-2.0-Pro)', v1.0, last update 31 July 2026, linked from https://seed.bytedance.com/transparency. Training-data collection date, not a stated knowledge cutoff. | ByteDance Seed | Not stated | Training data through | Not stated | -Not computed | lf3-static.bytednsdoc.com |
MiniMax-M3MiniMax-M3Note: EU AI Act Public Summary of Training Content for MiniMax-M3 (v1.0, 22/07/2026), linked from https://platform.minimax.io/docs/guides/transparency. Training-data collection date, not a stated knowledge cutoff. | MiniMax | Not stated | Training data through | -Not computed | file.cdn.minimax.io | |
Explainer
What a knowledge cutoff means
The dates above are the ones each provider publishes about its own model. Here is how to read them, and why they matter when you build on a model.
- What a knowledge cutoff is
- The date after which a model has seen little or no training data. Ask it about anything later and it either does not know or answers from older facts as if they were current. Providers state it on their model pages, in their own precision, from a month to an exact day.
- Reliable knowledge cutoff vs training data cutoff
- Anthropic publishes two dates for each Claude model. The reliable knowledge cutoff is the date through which the model's knowledge is most extensive and reliable; the training data cutoff is the end of the broader range of data it was trained on, so it can be later. For Claude Sonnet 4.5 they are Jan 2025 and Jul 2025. The reliable date is the safer one to plan around.
- Why recent facts still need retrieval or tools
- Models here ship a median of 5.5 months after their cutoff, then stay in production long after. New library versions, API changes and prices after the cutoff are unknown to the model. Put current facts in front of it: retrieval over your own docs (RAG), tool calls to live APIs, or the relevant documentation pasted into the prompt.
Not stated by the provider
99 models from 17 providers have no cutoff on the provider's model pages, docs or model cards as read on . Family-level or conflicting statements are not applied to a specific model.
Alibaba Cloud (Qwen)27
- Qwen-Flash
- Qwen-Max
- Qwen-MT-Flash
- Qwen-MT-Lite
- Qwen-Plus
- Qwen3-Coder-Flash
- Qwen3-VL-Plus
- Qwen3.5-122B-A10B
- Qwen3.5-27B
- Qwen3.5-35B-A3B
- Qwen3.5-397B-A17B
- Qwen3.5-Flash
- Qwen3.5-Omni-Flash
- Qwen3.5-Omni-Plus
- Qwen3.5-Plus
- Qwen3.6-27B
- Qwen3.6-35B-A3B
- Qwen3.6-Flash
- Qwen3.6-Plus
- Qwen3.7-Flash
- Qwen3.7-Max
- Qwen3.7-Plus
- Qwen3.8-2.4T-A95B
- Qwen3.8-27B
- Qwen3.8-Flash
- Qwen3.8-Max
- Qwen3.8-Omni-Flash
Z.ai13
Mistral AI8
xAI7
Baidu6
StepFun5
Tencent5
Google4
MiniMax4
Moonshot AI4
ByteDance Seed3
Cohere3
Inception3
AI21 Labs2
DeepSeek2
Xiaomi MiMo2
Amazon1
FAQ
Which AI model has the most recent knowledge cutoff?
Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1 and Claude Mythos 5.1 (Anthropic) have the newest provider-stated knowledge cutoff in this directory: Jun 2026. The newest from another provider is GPT-6 Luna (OpenAI) at May 18, 2026.
What is the difference between a reliable knowledge cutoff and a training data cutoff?
Anthropic lists both dates on each Claude model's overview page. They are the same month for 11 of the 16 Claude models here and differ for 5. The widest gap is Claude Sonnet 4.5: reliable knowledge cutoff Jan 2025, training data cutoff Jul 2025. Wherever the two differ, the training data cutoff is the later date.
How long after the knowledge cutoff is a model released?
Across the 62 models here with both a stated cutoff and a release date, the median gap is 5.5 months, ranging from 2 to 17 months.
Which AI providers do not publish a knowledge cutoff?
101 of the 167 models in the registry have no knowledge cutoff on their provider's pages. No knowledge cutoff is stated for any model from Alibaba Cloud (Qwen), Z.ai, Mistral AI, Baidu, StepFun, Tencent, MiniMax, Moonshot AI, ByteDance Seed, Inception, AI21 Labs, DeepSeek and Xiaomi MiMo. xAI, Google, Cohere and Amazon state one for some models but not others.
Verified from each provider's model pages, docs and model cards. Dates change when a provider updates a model or repoints an alias; check the linked source before you rely on one.
Newsletter
New models, every week
One email per week. New model releases, what changed and what it means for what you build.
Keep exploring
Model API pricing and specs
First-party prices per 1M tokens, context windows and sources for every model.
Model release timeline
Every dated API model release, newest first.
Model API capabilities
Tool calling, structured outputs, vision, caching and reasoning per model.
What is RAG?
How retrieval puts facts newer than the cutoff in front of a model.