Expired60% confidenceFactTime unknown
The Gemma 4 26B MOE version achieves an actual inference speed of 60 tokens per second on a GPU with 24GB of VRAM.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Gemma 4 Complete Guide: The Apache 2.0 Open-Source Agent Powerhouse
bilibiliAI全文总结
Related Claims
UnverifiedThe Gemma 4 26B MOE version can run on a graphics card with 24GB of VRAM.78% similarUnverifiedAfter quantization, Gemma 4 31B has a model size of approximately 20GB, with measured GPU memory usage of approximately 20.5GB and GPU utilization of approximately 91%.78% similarUnverifiedThe FP16 version of Gemma 3 12B requires approximately 24 GB of VRAM, while the Q4_K_M version reduces this to approximately 7–8 GB, with inference speed improving by approximately 2–3x and structured output accuracy losing only approximately 2–5%.77% similarUnverifiedThe Gemma E6B small model achieved a measured speed of over 46 tokens/second on that mini PC.74% similarUnverifiedGemma 4 31B can run on a regular consumer GPU with around 50 TOPS of compute.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/57145API
curl https://kongchang.com/api/v1/knowledge/claims/57145MCP
get_claim(id=57145)