Unverified50% confidenceBenchmarkExact time
On the 120-billion-parameter GPT-OSS 120B, token generation improves from 90 to 130 tokens/s, and PP improves from approximately 1,300 to approximately 3,200 tokens/s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/23/2026
First Seen
Valid until: 12/22/2026
Sources
Related Entities
Related Claims
UnverifiedReal-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).69% similarUnverifiedTaking Llama-3 70B as an example, the prefill stage for processing 32K tokens on an A100 80GB GPU takes approximately 800–1200ms, while processing 2K tokens takes only about 50ms.67% similarUnverifiedThe UP host estimates that inference speed could improve to 30–40 tokens/s after using a GPU with 50GB of VRAM (not confirmed through actual testing).67% similarUnverifiedA 200K token context request requires approximately 25× the KV-Cache memory compared to an 8K token request.66% similarUnverifiedH100的内存带宽约3.35TB/s，GPT-4规模模型单次生成一个Token需读取约1TB参数数据，理论上每秒最多生成约3个Token66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/944881API
curl https://kongchang.com/api/v1/knowledge/claims/944881MCP
get_claim(id=944881)