Snowflake Arctic Models Quiz

Answers and explanations appear as you take the quiz.

1. What kind of company built the Arctic models?
   • A cloud data-warehouse company
   • A video game studio
   • A car manufacturer
   • A social media network
2. What architecture figures did Snowflake report for Arctic?
   • A single dense network with 17B total parameters
   • 480B total parameters, with about 17B active per token
   • 480B active parameters for every token
   • A rule-based model with no neural parameters
3. What task focus distinguishes Arctic from general consumer chatbots?
   • Generating poetry and song lyrics
   • Playing chess
   • Enterprise tasks like SQL generation and coding
   • Editing photos
4. Under what license did Snowflake release Arctic’s model weights?
   • A closed proprietary license
   • A subscription-only license
   • No license at all
   • Apache 2.0
5. How did Snowflake characterize Arctic’s training compute budget?
   • It reported roughly under $2 million for its training compute
   • It reported a cost over one billion dollars
   • It said no computers were used
   • It did not disclose any estimate
6. What are the Arctic Embed models used for?
   • Generating images
   • Search, retrieval, and powering retrieval-augmented generation
   • Compressing video
   • Controlling robots
7. Which routing strategy did Snowflake report Arctic uses to choose MoE experts?
   • All-expert routing
   • Round-robin scheduling across tokens
   • Top-2 gating
   • A fixed expert selected by token position
8. Which two types of Arctic language-model checkpoints did Snowflake release?
   • Image and audio only
   • Encoder and speech-vocoder only
   • Training and evaluation datasets only
   • Base and instruct-tuned