Expired50% confidenceBenchmarkExact time
On the AIME-2024 mathematical reasoning benchmark, pass@1 improved from 0.39 to 0.49.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
Valid until: 10/10/2026(expired)
Sources
DeepSeek-V4 Flash RL Training Successfully Ported to AMD MI355X GPU: A Complete Breakdown
twitterlmsysorg7/10/2026
Related Claims
UnverifiedFor double-precision floating-point numbers, one ULP near the value 1.0 is approximately 2^-52 ≈ 2.22×10^-16.59% similarUnverifiedThe original Codex achieved a pass@1 score of 28.8% on the HumanEval benchmark, whereas using GPT-3 directly yielded only 0%.57% similarUnverified与INT4纯整数量化不同，浮点格式能更好地表示接近零的小数值57% similarUnverifiedIn XBEN white-box mode, T3MP3ST's pass@1 reached 98.7%, with the worst single round at 98.1% (102/104).56% similarUnverifiedGPTQ（2022年）和AWQ（2023年）能将70B参数的LLM量化到INT4精度，困惑度损失不到1%56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/490937API
curl https://kongchang.com/api/v1/knowledge/claims/490937MCP
get_claim(id=490937)