DeepSeek
Chinese AI lab with frontier-quality models at dramatically lower cost. DeepSeek-V3.2 (~1421 ELO on LMArena, ~1423 with thinking) offers near-frontier quality. V3.2 pricing: $0.28/$0.42 per 1M tokens (cache hit: $0.028 — 90% savings). Off-peak hours at 50% off. DeepSeek-R1 for reasoning at $0.12/$0.20 per 1M tokens. V4 expected mid-Feb 2026 with 1M+ token context via Sparse Attention and Engram memory for agentic tasks. Open-weight (V3 671B MoE, 37B active per token). 95% cheaper than GPT-5.
Overview
| Category | Ai |
| Self-Hostable | Yes |
| On-Prem | No |
| Best For | hobby, startup, growth |
| Last Verified | 2026-02-13 |
Strengths & Weaknesses
Strengths:- cost
- performance
- API hosted in China — latency and data sovereignty concerns for Western users
- Occasional capacity issues during peak demand
- Compliance certifications unclear for Western enterprises
- Content filtering aligned with Chinese regulations
- Chat web interface stability issues (OOM crashes after short usage)
When to Use
Best when:- Cost is the top priority — 10-20x cheaper than frontier proprietary
- Math, coding, or reasoning tasks
- Want to self-host an open frontier model
- Prototyping where data sensitivity is low
- High-volume production where cost matters
- Enterprise with strict compliance requirements
- Data cannot leave Chinese jurisdiction
- Need guaranteed uptime SLAs
- Sensitive content that may trigger content filters
Known Issues (1)
- [low] [BUG] Deepseek chat website OOM after short usage / 聊天网页短期使用后出现内存溢出（OOM）问题