Unverified50% confidenceBenchmarkExact time

在Agent's Last Exam、ZeroBench、DeepSWE三项测试中，V4-Flash-Vision-Exp成绩反超Opus 4.8

1

Sources

50%

Confidence

Short-term (~14 days)

Relevance

9/3/2026

First Seen

Valid until: 9/17/2026

Sources

DeepSeek V4首个多模态模型开源：305B权重MIT协议全放开

Related Entities

DeepSeek V4-Flash-Vision-Expmodel Opus 4.8model Agent's Last Examconcept ZeroBenchconcept DeepSWEconcept

Related Claims

Unverified盘古2.0 Flash在SWE-Bench Verified工程实战测试中得分仅63.1，在所有对比模型中垫底63% similar UnverifiedDeepSeek V4 Pro在Humanity's Last Exam基准测试中得分37.762% similar Unverified在HLE科学类基准上，Agent A1 据实测数据超过了Qwen 3.6和Step 3.5 Flash62% similar UnverifiedOpus 5在AI Omniscience测试拿了31分，避免瞎编的能力优于Opus 4.862% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/849885

API

curl https://kongchang.com/api/v1/knowledge/claims/849885

MCP

get_claim(id=849885)