Unverified50% confidenceFactExact time
Under a single-request serial load, each request decodes at 648–649 tokens/s, and the whole server produces 589–593 tokens/s.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen the context approaches 512K tokens, the prefill-decode disaggregated architecture ensures that users still receive the first response token within 800 milliseconds after reading an entire specification.74% similarUnverifiedDeveloper Primitive reported: 91.2 tokens/s with speculative decoding off, reaching 142.6 tokens/s after enabling 3 speculative tokens, approximately a 56% speedup under the same configuration.69% similarUnverifiedA log study measured the tokens each harness sends before reading the user's prompt: OpenCode approximately 7,000 tokens, CloudCode approximately 33,000 tokens, and Pi approximately 1,300 tokens.68% similarUnverifiedThrough pure CPU inference and the 4-bit GGUF quantization format, llama.cpp enables a 7B-parameter model to achieve a generation speed of 20–40 tokens per second on an ordinary server without a GPU.67% similarUnverifiedReal-world testing with the 284-billion-parameter DeepSeek V4 Flash shows token generation rising from 37 to 53 tokens/s (approximately 1.5x), and Prompt Processing jumping from 483 to 1,485 tokens/s (approximately three times).67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1014545API
curl https://kongchang.com/api/v1/knowledge/claims/1014545MCP
get_claim(id=1014545)