AI models go live a median of 4 months after their knowledge ends: Claude Opus 5.5 knows up to June 2026, Gemini 4 Argon only to March

Of the 22 live AI models whose labs publish a knowledge cutoff, the median went live 4 months after its training data ends, by ArtificialWatch's count. The latest published cutoff we found is June 2026, for Anthropic's four current Claude models and OpenAI's GPT-6.1 Terra.

THE LATEST CUTOFFS

• June 2026: Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 (Anthropic gives the month, for both its reliable-knowledge and its training-data cutoff), and GPT-6.1 Terra (June 1, 2026).
• May 2026: GPT-6 Luna (May 18), Grok 4.7 and Claude Opus 5.
• April 2026: GPT-6.1 Sol and GPT-6 Astra (both April 30) and GPT-6 Sol (April 20).
• March 2026: Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite.

FIVE GOOGLE MODELS, ONE CUTOFF

Google DeepMind's model cards give all five of its current Gemini models a March 2026 cutoff. They went live over 12 weeks: 3.5 Flash-Lite and 3.6 Flash on July 21, 3.7 Flash on August 13, 3.8 Flash on September 2 and Gemini 4 Argon on October 13. So the later each shipped, the further its knowledge lagged: 4 months for the July pair, 5 for 3.7 Flash, 6 for 3.8 Flash and 7 for Argon, the longest gap on the list. The cards say 3.8 Flash is based on 3.7 Flash, and 3.7 Flash on 3.6 Flash.

THE FLAGSHIPS

• Claude Opus 5.5: June 2026 cutoff, live September 22. 3 months.
• Grok 4.7: May 2026, live September 21. 4 months.
• GPT-6.1 Sol: April 30, 2026, live September 29. 5 months. GPT-6 Astra, live September 3, has the same April 30 cutoff.
• Gemini 4 Argon: March 2026, live October 13. 7 months.

The shortest gap is Claude Opus 5's: a May 2026 cutoff and a July 24 launch, 2 months.

HOW FAST THE CUTOFFS MOVE

Anthropic's Claude Opus 4.8 (live May 27) and the three models after it, Mythos 5, Claude Sonnet 5 and Fable 5 (June 9 to July 1), all stop in January 2026. Claude Opus 5 moved to May, and Fable 5.1 (September 1) and the three Claude 5.5 models to June. OpenAI's went from December 1, 2025 for GPT-5.5 (April 24) to February 16 for ChatGPT-5.6 (July 9), April 30 for GPT-6 Astra and June 1 for GPT-6.1 Terra (October 26).

WHO PUBLISHES NONE

We found no cutoff on the pages we read for DeepSeek (its API docs and the DeepSeek V4.1 Flash card on Hugging Face), Moonshot (the Kimi K3.1 Pro docs and Hugging Face card), Zhipu AI (the GLM-5.3 docs and Hugging Face card), Mistral (the Mistral Large 4 docs) or MiniMax (its model docs). xAI's models page states one for Grok 4.7 only. Model catalogs fill some of these gaps with dates the labs don't state, and they can contradict the labs: models.dev lists August 2025 for Claude Sonnet 4.6, where Anthropic's transparency hub says May 2025. So the table uses only dates read on the lab's own page.

WHAT WE DO NOT KNOW

A cutoff given as a month could fall anywhere in it, so a gap can be off by up to a month either way. Labs don't define the date the same way: Anthropic separates a reliable-knowledge cutoff from a training-data cutoff, and Google says a model's knowledge may be less up to date in some domains. A model updated after launch can keep the cutoff it shipped with.

HOW WE COUNTED

Each cutoff was read on November 6 on the lab's own page: Anthropic's models overview and transparency hub, OpenAI's model pages, Google DeepMind's model cards and xAI's models page. The release date is the day ArtificialWatch verified the model live on a public API, from its release log. The gap is the number of calendar months from the cutoff's month to the release month, so a June cutoff and a September launch count as 3. The full table, with a link to each lab's page, is at artificialwatch.com/llm-knowledge-cutoffs.