Unverified50% confidenceBenchmarkExact time
The dual-DGX Spark setup prefills at about 2,100 tokens/s, with a first-token wait of about 63 seconds for a 128K prompt; a single unit has 273GB/s of bandwidth, and with DraftHead enabled it achieves 87 tokens/s written in coding scenarios and 131 tokens/s for structured output.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/31/2026
First Seen
Valid until: 1/29/2027
Sources
Related Entities
Related Claims
UnverifiedIt is generally considered that single-concurrency output above 100 tokens/second is sufficient to achieve a smooth typewriter effect, and at approximately 200 tokens/second, generating a 2,000-character long text takes only about 10 seconds.73% similarUnverifiedIn the TP4 configuration on 4 DGX Spark units, measured TTFT is: about 12 seconds for over 60,000 tokens (about 5400 Token/s), about 54 seconds for 260,000 tokens (about 4800 Token/s), and about 4 minutes 54 seconds for 1 million tokens (about 3400 Token/s).73% similarUnverifiedWith 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.71% similarUnverifiedIn a test scenario where eight agents each process 7.5k-token documents and a chat request is issued in the second second, Superfluid answers in 0.36 seconds while llama-server outputs no first token for more than 10 seconds, a first-token latency gap of about 30 times.69% similarUnverifiedIn the test environment, MiniMax M3.1 Flash's output throughput held steady at about 112 tokens per second, and a single 6,000-token think takes close to a minute of wall-clock time.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1121823API
curl https://kongchang.com/api/v1/knowledge/claims/1121823MCP
get_claim(id=1121823)