DeepSeek: DeepSeek V3.1 Terminus wins on more metrics (6 of 9), but the right pick depends on what you optimize for — see the breakdown below.
| Metric | DeepSeek: DeepSeek V3.1 Terminus | Qwen: Qwen3 VL 235B A22B Thinking |
|---|---|---|
| Intelligence Index | 28.5✓ | 27.6 |
| Coding Index | 31.9✓ | 20.9 |
| GPQA Diamond | 75% | 77%✓ |
| Design Arena Elo | 1238✓ | — |
| Speed (tokens/sec) | 26 | 45✓ |
| Latency | 933ms✓ | 1.0s |
| Input price /M | $0.270 | $0.260✓ |
| Output price /M | $0.950✓ | $2.60 |
| Context window | 164K✓ | 131K |
| Capabilities | ReasoningToolsJSON | ReasoningToolsJSONVision |