Make the trade-off visible.
Choose up to three models. Compare the same operation and billing unit, then validate quality on your own prompts.
These are published gateway facts, not an independent benchmark. Model quality and response time depend on the task and are not ranked here.
A starting point from the current chat catalogue
| Attribute | deepseek-aiDeepSeek-V4-Flash-0731 Explore model ↗ | deepseek-aiDeepSeek-V4-Flash Explore model ↗ | deepseek-aiDeepSeek-V4-Flash-Vision-Exp Explore model ↗ |
|---|---|---|---|
| Availability | Available | Available | Available |
| Operation | Text chat | Text chat | Text chat |
| Gateway context limit | 1,048,576 | 1,048,576 | 1,048,576 |
| Maximum output | 4,096 | 4,096 | 4,096 |
| Input · IRT / 1M tokens | 17,643 | 26,464 | 129,376 |
| Output · IRT / 1M tokens | 52,927 | 52,927 | 388,128 |
| Published unit ceiling | Not listed | Not listed | Not listed |
| Tool calling | Supported | Supported | Supported |
| Streaming | Supported | Supported | Supported |
| Reasoning | Supported | Supported | Supported |
How to read this comparison
- Availability describes the gateway’s current sale checks, not a guarantee of capacity or uptime.
- Context and output figures are gateway limits. The underlying model may have a different published maximum.
- A blank rate or capability is not a zero price or a promise of support.
- Compare output quality with your own test data; no synthetic benchmark scores are assigned.