Unverified80% confidenceOpinionTime unknown
3B参数模型在SVG生成任务上存在能力天花板，量化级别的影响被基础能力不足所掩盖
1
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在纯SVG等角投影微型城市绘制任务中，DeepSeek V4 Pro获得4/10分，存在裁剪问题和比例失衡62% similarUnverified在SVG蝴蝶图形测试中，Qwen3经过两次迭代即达到高质量输出61% similarUnverified在前端与视觉任务（Three.js、SVG等）中Sol仅得6至7分，而Fable 5能稳定拿到9至10分，前端仍是OpenAI的弱项60% similarUnverifiedMuse Glimmer 30B在SVG生成任务中表现最差，生成的猫角度诡异，围栏和太阳粗糙，评测者给SVG能力打了最低分59% similarUnverified响应时间只看厂商标称的1ms/0.5ms GtG意义不大，应看RTINGS等第三方实测的过冲(overshoot)数据，标称与实测常差2-3倍59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/34085API
curl https://kongchang.com/api/v1/knowledge/claims/34085MCP
get_claim(id=34085)