Unverified50% confidenceBenchmarkExact time
Ziskind tested DeepSeek V4 Flash with a 32,000-token codebase: the 256GB M5 Ultra took about 50 seconds to first token, while two DGX Spark units totaling about $10,000 took only 17 seconds.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/14/2026
First Seen
Valid until: 2/12/2027
Sources
Related Entities
Related Claims
UnverifiedBased 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.72% similarUnverifiedLink 3.0 Flash has released its full weights, and in tests on a single Nvidia DGX Spark using 1319 GSM8K questions, decoding throughput improved by 42.5%, average latency dropped from 15.21 seconds to 11.49 seconds, and decoding speed rose from 37.81 tokens per second to 53.89.72% similarUnverifiedWith tensor parallelism across four M5 Ultras, Muse Spark Open's decoding speed rose from 15.74 tokens/s under two-machine pipeline parallelism to 36.73 tokens/s, and Qwen 3.8 rose from 19.44 to 50.89 tokens/s.70% similarUnverifiedZiskind tested DeepSeek V4 Flash concurrency: the Mac's total throughput peaked at four users and declined at eight users; with eight users and chat history, each person waited over a minute to first token, after which generation was about 5 tokens per second.68% similarVerifiedRunning Muse Glimmer at full precision requires 55 GB of VRAM, which is compressed to under 20 GB via 4-bit quantization, allowing it to run on consumer-grade GPUs with 24 GB of VRAM.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1169141API
curl https://kongchang.com/api/v1/knowledge/claims/1169141MCP
get_claim(id=1169141)