Nemotron 3.5 Super 120B (Reasoning) vs DeepSeek V4 Flash (high)
Results updated 9 November 2026 · In Terminus-2 and Claude Code, so part of any gap can come from the harness · How we benchmark
Verdict
DeepSeek V4 Flash (high) beats Nemotron 3.5 Super 120B (Reasoning) by 3 points, and is faster; Nemotron 3.5 Super 120B (Reasoning) is cheaper.
- Cost per task
- $0.011 vs $0.080
- Time per task
- 13.8 min vs 5.4 min
- Tokens per attempt
- 105k vs 634k
- Image input
- No vs No
Nemotron 3.5 Super 120B (Reasoning), Terminus-2 scores higher on GeoBenchX; DeepSeek V4 Flash (high), Claude Code scores higher on Earth-Bench V2 and GS-QA2. Nemotron 3.5 Super 120B (Reasoning), Terminus-2 is cheaper ($0.011 against $0.080 per task), but DeepSeek V4 Flash (high), Claude Code is faster (5.4 min against 13.8 min).
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Nemotron 3.5 Super 120B (Reasoning) | DeepSeek V4 Flash (high) |
|---|---|---|
| Remote sensing and raster analysis | 54 | 72 |
| GIS analysis | 70 | 63 |
| Climate and time series | 50 | 56 |
| Making maps | 70 | 74 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Nemotron 3.5 Super 120B (Reasoning), Terminus-2 · DeepSeek V4 Flash (high), Claude Code.
Side by side
| Metric | Nemotron 3.5 Super 120B (Reasoning), Terminus-2 | DeepSeek V4 Flash (high), Claude Code |
|---|---|---|
| Geospatial Agent Index: Text only | 56 | 59 |
| Geospatial Agent Index: Text + vision | No data | No image input |
| Earth-Bench V2 score | 49 | 56 |
| GeoBenchX score | 56 | 53 |
| GS-QA2 score | 63 | 69 |
| Vision agents score | No data | No image input |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 83 | 90 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 79 | 89 |
| OGD4All score (standalone; in no index) | 85 | 77 |
| GeoNatureAgent score (standalone; in no index) | 100 | 71 |
| MultiGlobeQA score (standalone; in no index) | 70 | 67 |
| GeoAgentEval score (standalone; in no index) | 84 | 80 |
| ImpactMesh score (standalone; in no index) | No data | No data |
| Remote sensing and raster analysis score | 54 | 72 |
| Change detection score | 56 | 56 |
| GIS analysis score | 70 | 63 |
| Climate and time series score | 50 | 56 |
| Making maps score | 70 | 74 |
| Cost per task | $0.011 | $0.080 |
| Time per task (at 100 output tokens/s) | 13.8 min | 5.4 min |
| Tokens per attempt | 105k | 634k |
| Turns per attempt | 10.5 | 16.1 |
| Input price per 1M tokens | $0.20 | $0.15 |
| Output price per 1M tokens | $0.50 | $0.60 |
| Context window | 262,144 | 1,048,576 |
| Image input | No | No |
| Developer | NVIDIA | DeepSeek |