DeepSeek
62.6
As usual
experience score
▲ 0.7
Confidence: Medium
501 opinions tracked this week · 7-day rolling window
30-day experience score
Version breakdown
Only comments that explicitly name a version contribute to its score.
| DeepSeek V4 Flash | 61.0−1.6 | n=148 | Confidence: Medium | |
|---|---|---|---|---|
| DeepSeek V4 Pro | 66.2+3.5 | n=31 | Confidence: Medium | |
| DeepSeek V3 | 52.9−9.7 | n=10 | Confidence: Low | |
| DeepSeek V3.2 | mentions only | n=2 | Confidence: Insufficient | |
| R1 | mentions only | n=2 | Confidence: Insufficient |
305 generic family mentions are included in the family score, not any version score.
By experience dimension
| General text | 65 | n=181 | |
|---|---|---|---|
| Coding | 62 | n=62 | |
| Reasoning | 64 | n=74 | |
| Image / vision | 40 | n=8 | |
| Roleplay / creative | 48 | n=14 | |
| Safety & refusals | not enough data | n=1 | |
| Speed & latency | 62 | n=136 | |
| Local deploy | 53 | n=25 |
Top positive drivers
- speed quality Speed & latency ×61
- strong reasoning Reasoning ×37
- competitor flight General text ×30
- pricing billing General text ×22
- problem solved General text ×15
Top negative drivers
- performance issue Speed & latency ×23
- speed quality Speed & latency ×12
- competitor flight General text ×10
- hallucination Reasoning ×9
- performance issue Coding ×8
What users are saying
🔴 “never done post-training well”
via zhihu · text_general · zh
🔴 “前提是ds愿意搞后训练，能搞后训练，讲真现在的水平对不起这1.6T的模型”
via zhihu · text_general · zh
🔴 “DeepSeek是通用性模型，不专注于编程；Copilot说是反应用，其实还是主要在编程上。”
via zhihu · coding · zh
🔴 “I would probably recommend something other than Deepseek unless you've got the RAM for cpu-offload.”
via reddit · local_deploy · en
🔴 “为什么更新后，deepseek语言风格让我人很难受？”
via zhihu · roleplay_creative · zh