Unverified50% confidenceBenchmarkExact time
The 2B-parameter Gemini Nano model is nearly on par with last year's 27B model on the MMLU Pro benchmark.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/29/2026
First Seen
Valid until: 1/27/2027
Sources
Android ML Kit On-Device AI: Performance Optimization and Quality Breakthroughs
youtubeAndroid Developers10/29/2026
Related Entities
Related Claims
UnverifiedGemini Nano的参数规模约为18亿（Nano-1）和32.5亿（Nano-2）68% similarUnverifiedThe first 20 layers of the Gemini Nano model share the KV cache and support a context length of 16K tokens.66% similarUnverifiedThere are rumors that the new Gemini 4 Pro model may support a context length of up to 2M tokens.66% similarUnverifiedGoogle's Gemini Nano is designed for on-device deployment with a parameter scale of approximately 1.8B–3.2B.65% similarUnverifiedGemini Nano was distilled from the Gemini family of large models, with its parameter count kept in the range of 1 billion to 3 billion.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1112922API
curl https://kongchang.com/api/v1/knowledge/claims/1112922MCP
get_claim(id=1112922)