DeepSeek V3
Last updated: August 10, 2026
Specs
| Vendor | DeepSeek |
|---|
| Released | December 26, 2024 |
|---|
| Context window | 128K tokens |
|---|
| Input | $0.26 / M tokens |
|---|
| Output | $1.03 / M tokens |
|---|
Benchmarks
| mmlu | 88.5 |
|---|
| gpqa | 59.1 |
|---|
| humanEval | 86 |
|---|
| arenaElo | 1355 |
|---|
| sweBench | 42 |
|---|
| sweBenchPro | 32 |
|---|
| liveCodeBench | 60 |
|---|
| aiderPolyglot | 49.6 |
|---|
| terminalBench | 22 |
|---|
| mmluPro | 75.9 |
|---|
| hle | 5 |
|---|
| aime | 59.4 |
|---|
| math500 | 90.2 |
|---|
Strengths
- Very low price with GPT-4-class chat quality
- Open weights, self-host on your own hardware
- Strong code generation for its price
Weaknesses
- Not a reasoning-first model (use R1 for that)
- No native multimodal
- Tool use less polished than OpenAI/Anthropic
Best for
- High-volume chat with tight unit economics
- Self-hosted chat deployments
- Content generation at scale
Compare with other AI models