Unverified50% confidenceBenchmarkExact time
When enabling multi-token prediction (MTP) with a draft depth of 2, the R9700 rose from 27.1 to 46.2 tokens/s (+70%) and the B70 rose from 21.4 to 28.4 tokens/s (+33%).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
UnverifiedQwen 3.8 has a base speed of 22 tokens/s on the M5 Max, which can be boosted to 34 or even 56 tokens/s with MTP enabled.71% similarUnverifiedIn speculative sampling, when the acceptance rate of candidate tokens is high, the effective number of output tokens per forward pass can reach 2–4, thereby improving the overall generation speed by 1.5x–3x.68% similarUnverifiedAccording to Artificial Analysis testing, Grok 4.6 consumes approximately 38k tokens per task at the MXX High tier, while Grok 4.7 surges to 81k tokens.68% similarUnverifiedIn token generation tests for the Gemma 4 12B model, Spark reached 26.4 tokens/s, Halo 24.6 tokens/s, and M4 Pro was the fastest at 33.8 tokens/s.67% similarUnverifiedAfter expanding the context from 256 tokens to 2,048 tokens, the evidence gain under fixed-answer conditions on FEVEROUS improved by 3.84 (Qwen) and 3.10 (Llama) percentage points respectively.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977978API
curl https://kongchang.com/api/v1/knowledge/claims/977978MCP
get_claim(id=977978)