DeepSeek V4 Pro: what developers actually say
We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on DeepSeek V4 Pro. 104 of the 121 opinions come from people who say they use it themselves.
Developers give DeepSeek V4 Pro 75% approval across 121 opinions from 85 authors, and 104 of those opinions come from firsthand use. Price drives the positive view: cost and pricing draw 55 positive mentions against 3 negative. Code quality is split at 31 positive and 10 negative, and reliability is the weak spot at 9 negative against 2 positive. Approval over the last 90 days is 67%, below the overall figure, and monthly approval moved from 77% in May to 84% in June to 69% in July, while August and September hold only 7 and 5 opinions, too few to read as a trend.
Pick it if
- Want low cost for heavy daily use
- Are building hobby or side projects on a budget
- Want generous usage limits
Look elsewhere if
- Need reliable multi-step tool use
- Cannot absorb rework time
- Need top-tier code quality
Strengths and weaknesses
Each row counts opinions that mentioned the topic. Bars to the right of the line are praise, bars to the left are complaints.
How opinion has moved
What developers say
Praise
Complaints
Summaries are paraphrased. Each links to the original comment.
DeepSeek V4 Pro compared
FAQ
Is DeepSeek V4 Pro worth it?
For cost-sensitive work, the data supports yes. It holds 75% approval, and cost and pricing get 55 positive mentions against 3 negative. Users report about $24 a month during a promo and under $100 a month for near-exclusive use.
Is DeepSeek V4 Pro better than GLM 5.2?
The data leans no. GLM 5.2 was named as the preferred alternative 4 times, more than any other model. One user also ranked DeepSeek V4 Pro consistently worse than GLM 5.2 at Max reasoning.
Is DeepSeek V4 Pro reliable for coding?
Reliability is the weak area. It draws 9 negative mentions against 2 positive. Reports mention broken changes that need reverting, hallucinated problems, and 30 to 60% extra time to fix output.