Unverified50% confidenceBenchmarkExact time
With thinking mode off and 16 streams combined, code task generation speed is 369.9 Token/s and prose is 317.8 Token/s; from a single-stream perspective, code is 119.8 Token/s and prose is 71.2 Token/s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
4 DGX Sparks Test DeepSeek V4.1 Flash: How Good Is Local Inference Productivity?
bilibiliAI一平米10/20/2026
Related Entities
Related Claims
UnverifiedLong context has little effect on generation speed: after 4K context the median time per output token is 7.6 ms, and after 256K it is 8.3 ms; but the first read of a 128K-token prompt takes 18.33 seconds, so prefix caching is critical.75% similarVerifiedIn decode speed tests, speeds of approximately 45 tokens/second are achievable within 20,000 tokens, dropping sharply to approximately 20 tokens/second beyond 30,000 tokens.72% similarUnverifiedIn this test, the prefill speed was fixed at approximately 500–600 tokens/s, and the time to first token increased linearly with input length.70% similarUnverifiedPerplexity published self-tested latency (September 30): under 2 seconds for a few hundred input tokens, about 5 seconds for 90k tokens, about 14 seconds for 190k tokens, and about 23 seconds near the 262k limit.70% similarUnverifiedLocal model inference speed is approximately 1–10 tokens/second, while cloud-based APIs can reach 50–150 tokens/second.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1077055API
curl https://kongchang.com/api/v1/knowledge/claims/1077055MCP
get_claim(id=1077055)