DeepSeek
DeepSeek V4
DeepSeek's 2026 open-weight model trained for million-token contexts with a multi-teacher distillation pipeline.
Our editorial assessment as of Jun 4, 2026, based on the public sources listed below.
Overall rank
#24
Tier
A
Score
81.0
Main category
Overall
Strengths
- Open-weight with permissive licensing
- Very low API pricing among frontier-class models
- Trained on 32T+ tokens with domain-specialist experts
Considerations
- Trails the very top closed frontier models on hardest evals
- Self-hosting requires substantial GPU planning
- Detailed public benchmarks still limited
Price
Open weights (free) · API ~$2.2/1M tokens
Score breakdown
Performance
82.0
Reputation
80.0
Price
80.0
Recency
90.0
Verification status
Each field carries its own re-verification date. Amber means older than 90 days — treat those figures as stale until re-checked.
Sources
- AI Benchmarks 2026: Top Evaluations and Their LimitsAccessed: 2026-05-13
- Artificial Analysis: AI Model & API Providers AnalysisAccessed: 2026-05-13