Unverified50% confidenceBenchmarkExact time
On the same Google Cloud single-card T4, Gemma 4 E2B decodes at 115.3 tokens/s when cold at 45°C and drops to 106.8 tokens/s when hot at 77°C, a sustained speed drop of about 7.4%.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLing 3.0 Tiny achieves an inference speed of approximately 86–90 tokens per second on a MacBook with M4 Pro, with a peak memory usage of approximately 8.34 GiB under an 8K context.68% similarUnverifiedIn short-answer scenarios, on the same M4 Mac, merely changing the power mode jumps the speed from 28.5 tokens/s to 109.6 tokens/s, a difference of 3.84 times.64% similarUnverifiedOn 6 RTX 5090 GPUs, Colibri's measured speed reaches 9.0–9.2 tokens/s, while on a 25GB development machine streaming from NVMe the speed is 0.05–0.1 tokens/s.62% similarUnverifiedBased on public data for high power plus MLX or Ollama, the decoding speed of different models on M4 Max is roughly a ladder as follows: Nemotron 3 Nano about 160 tokens/s, Qwen 3.5 35B at 107 to 130 tokens/s, Gemma 4 26B about 75 tokens/s, DeepSeek V4 Flash about 27 tokens/s, and dense 27B about 22 tokens/s.62% similarUnverifiedWhen the MTP draft depth is raised from the default 4 to 6 or 8, Qwen3.8-27B's output speed drops to 53.9 and 52.4 tokens/second respectively, below the default 55.1 tokens/second.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047183API
curl https://kongchang.com/api/v1/knowledge/claims/1047183MCP
get_claim(id=1047183)