Unverified50% confidenceBenchmarkExact time
In 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.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
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.77% similarUnverifiedAccording to Stridenote, Ollama tests showed that Qwen3.5 35b's speed jumped from 58 token/s to 112 token/s after switching to Apple Silicon's MLX.73% similarUnverifiedIn benchmark testing of prefilling a 2000-token file, the speed under a 16k context increased from 109 tok/s on Mac alone to 157 tok/s on Mac+iPhone (+44%).69% 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.69% similarUnverifiedOn the RTX 5090, using D-Flash speculative decoding can achieve a speedup of more than 3x, while the improvement on Mac M4/M5 is between 1.5x and 1.8x.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1015147API
curl https://kongchang.com/api/v1/knowledge/claims/1015147MCP
get_claim(id=1015147)