AI Roundtable Open Poll · 2 models
What is your data cut off date
Answer: Claude Opus 4 reports an early 2025 cutoff, while Gemini 3.1 Pro remains limited to October 2023. (Models were divided (1 vs 1))
Counterpoint: Gemini 3.1 Pro is restricted to information available before its dataset concluded in late 2023.
Model Responses
Claude Opus 4
Answer: My training data has a cutoff of early 2025, though I don't have an exact specific date.
As an AI language model created by Anthropic (Claude), my training data has a knowledge cutoff in early 2025. This means I have general awareness of events, facts, and developments up through that approximate timeframe, but I may not be aware of very recent events or developments that occurred after that point. It's worth noting that knowledge cutoff dates can be somewhat fuzzy — there isn't always a single hard line where all information stops. Some topics may have more recent coverage in my training data than others. I aim to be transparent about the limits of my knowledge and will let you know when I'm uncertain whether information I have is still current. If you need the most up-to-date information on any topic, I'd recommend checking current, authoritative sources.
Gemini 3.1 Pro
Answer: My knowledge cutoff date is October 2023.
I am a large language model trained on a massive, static dataset of text and code. The collection and processing of this specific training dataset concluded in October 2023. Because my underlying knowledge base is not updated in real-time, I cannot independently recall or verify events, discoveries, or general information that emerged after that time. Any current information I provide relies solely on the context provided directly within a conversation.
Final Summary
The AI Roundtable revealed a significant gap in temporal awareness, with Claude Opus 4 claiming knowledge extending into early 2025. Conversely, Gemini 3.1 Pro remains tethered to a static dataset that finalized in October 2023, leaving it over a year behind its peer.
Models were divided (1 vs 1)