Overview
Geospatial Intelligence IndexNot yet rankedText only · Higher is better
Geospatial Agent IndexNot yet rankedText only · Higher is better
Cost per taskNot yet rankedAgent tasks, at list prices · Lower is better
Time per taskNot yet rankedAt 100 output tokens/s · Lower is better
Technical specification
- Developer
- DeepSeek
- Released
- 10 September 2026
- Weights
- Open
- Harness
- Terminus-2 2.0.0
- Reasoning setting
- max
- Context window
- 1,048,576 tokens
- Maximum output tokens
- 65,536
- Image input
- Yes
- Temperature
- 0.6
- Input price per 1M tokens
- $0.15
- Cached input price per 1M tokens
- $0.0030
- Output price per 1M tokens
- $0.60
- Map and trace judge
- Claude Sonnet 5.5
Frequently asked questions
Is DeepSeek V4.1 Flash good at spatial reasoning?
We benchmark DeepSeek V4.1 Flash on geospatial agent tasks (GIS and Earth observation work with real data). Its best result is 76 out of 100 on the Geospatial Agent Index, 16th of 92 entries, at max reasoning in Codex. See that entry or the leaderboard.
Can DeepSeek V4.1 Flash analyse satellite or map images?
It takes images, but we have not benchmarked it on image tasks yet.
How does DeepSeek V4.1 Flash compare with Claude Opus 4.8?
Its closest rival on the Geospatial Agent Index is Claude Opus 4.8 at 76, so DeepSeek V4.1 Flash (76) is level, within the noise of about 3 points. Compare them side by side.