gpt-oss-120b (high) vs DeepSeek V4 Flash (high)
Results updated 30 October 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) clearly beats gpt-oss-120b (high) by 10 points; gpt-oss-120b (high) is cheaper and faster.
- Cost per task
- $0.026 vs $0.080
- Time per task
- 3.0 min vs 5.4 min
- Tokens per attempt
- 57k vs 634k
- Image input
- No vs No
DeepSeek V4 Flash (high), Claude Code scores higher on GeoBenchX and GS-QA2; they are level on Earth-Bench V2. gpt-oss-120b (high), Terminus-2 is cheaper and faster: $0.026 against $0.080 per task, and 3.0 min against 5.4 min.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | gpt-oss-120b (high) | DeepSeek V4 Flash (high) |
|---|---|---|
| Making maps | 40 | 68 |
| Remote sensing and raster analysis | 44 | 67 |
| Recognising infeasible requests | 58 | 35 |
| Spatial statistics | 15 | 38 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: gpt-oss-120b (high), Terminus-2 · DeepSeek V4 Flash (high), Claude Code.
Side by side
| Metric | gpt-oss-120b (high), Terminus-2 | DeepSeek V4 Flash (high), Claude Code |
|---|---|---|
| Geospatial Agent Index: Text only | 46 | 56 |
| Geospatial Agent Index: Text + vision | No image input | No image input |
| Earth-Bench V2 score | 50 | 50 |
| GeoBenchX score | 45 | 50 |
| GS-QA2 score | 42 | 69 |
| Vision agents score | No image input | No image input |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 70 | 84 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 53 | 89 |
| OGD4All score (standalone; in no index) | No data | 70 |
| GeoNatureAgent score (standalone; in no index) | No data | 71 |
| MultiGlobeQA score (standalone; in no index) | No data | 64 |
| GeoAgentEval score (standalone; in no index) | No data | 80 |
| Remote sensing and raster analysis score | 44 | 67 |
| GIS analysis score | 41 | 62 |
| Climate and time series score | 53 | 53 |
| Spatial statistics score | 15 | 38 |
| Making maps score | 40 | 68 |
| Recognising infeasible requests score | 58 | 35 |
| Cost per task | $0.026 | $0.080 |
| Time per task (at 100 output tokens/s) | 3.0 min | 5.4 min |
| Tokens per attempt | 57k | 634k |
| Turns per attempt | 5.8 | 16.1 |
| Input price per 1M tokens | $0.35 | $0.44 |
| Output price per 1M tokens | $0.75 | $1.32 |
| Context window | 128,000 | 1,048,576 |
| Image input | No | No |
| Developer | OpenAI | DeepSeek |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, gpt-oss-120b (high), Terminus-2 or DeepSeek V4 Flash (high), Claude Code?
DeepSeek V4 Flash (high), Claude Code, 56 to 46.
Which is cheaper per task, gpt-oss-120b (high), Terminus-2 or DeepSeek V4 Flash (high), Claude Code?
gpt-oss-120b (high), Terminus-2, $0.026 to $0.080.
Which writes fewer output tokens per task, gpt-oss-120b (high), Terminus-2 or DeepSeek V4 Flash (high), Claude Code?
gpt-oss-120b (high), Terminus-2, 15k to 19k.
Which has the larger context window, gpt-oss-120b (high), Terminus-2 or DeepSeek V4 Flash (high), Claude Code?
DeepSeek V4 Flash (high), Claude Code, 1,048,576 tokens to 128,000 tokens.