Unverified50% confidenceBenchmarkExact time
腾讯官方自测数据宣称混元3几乎在每个领域都超过DeepSeek-V4-Pro，仅MCP相关测试项略逊
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/10/2026
First Seen
Valid until: 10/8/2026
Sources
腾讯混元3开源：295B MoE架构全解析，256K上下文成最大短板
bilibili鲲鹏Talk7/6/2026
Related Claims
UnverifiedPhi-3-mini（3.8B参数）在部分基准测试上超越了Mixtral 8x7B等更大的混合专家模型66% similarUnverifiedKimi K3的2.8万亿参数规模超越了此前最大的开放权重模型DeepSeek-V3的6710亿参数65% similarUnverifiedGLM-4.7在代码能力基准测试中超过DeepSeek V3.2及Claude Sonnet 4.565% similarUnverifiedQwen3-4B在多项推理类评测指标上已与DeepSeek V3或GPT-4o相差无几65% similarUnverifiedDeepSeek-V3发布时在多项基准测试中与Claude 3.5 Sonnet持平甚至超越，训练成本据报道仅为主流闭源模型的数分之一64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/465611API
curl https://kongchang.com/api/v1/knowledge/claims/465611MCP
get_claim(id=465611)