Unverified50% confidenceBenchmarkExact time
Deep-SWE是更严格的新基准，DeepSeek-V4官方分约8%，Rocky接入测20题目前得分为0
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/25/2026
First Seen
Valid until: 10/23/2026
Sources
Related Claims
UnverifiedDeepSeek目前只有约两万张H系列等效GPU算力67% similarUnverifiedDeepSeek的编程智能体DeepSWE在SWE-bench基准测试中从预览版的12.8分提升至62.7分67% similarUnverifiedDeepSeek-V4-Pro has not yet surpassed the benchmark scores of top competitors as of now, and some benchmark scores are still estimated values.67% similarUnverifiedDeepSeek offers extremely high cost-effectiveness, with very low prices; DeepSeek-Coder's performance on SWE-bench is close to GPT-4 level, yet its price is only one-tenth or even lower.66% similarUnverifiedDeepSeek V4 Pro's code engineering score of 62.7 is 7.3 points below Fable 5, its data science hard-level score of 67.2 trails Opus 4.8 by approximately 4.5 points, and its high-difficulty reasoning score of 60 ranks second globally.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/615557API
curl https://kongchang.com/api/v1/knowledge/claims/615557MCP
get_claim(id=615557)