Nemotron 3.5 Bolt E4B (Reasoning) vs gpt-oss-20b (high)
Results updated 9 November 2026 · In Terminus-2 and Codex, so part of any gap can come from the harness · How we benchmark
Verdict
gpt-oss-20b (high) beats Nemotron 3.5 Bolt E4B (Reasoning) by 5 points, and is faster.
- Cost per task
- No data vs $0.094
- Time per task
- 5.2 min vs 4.8 min
- Tokens per attempt
- 341k vs 457k
- Image input
- No vs No
gpt-oss-20b (high), Codex scores higher than Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 on all 3 evaluations scored for both.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Nemotron 3.5 Bolt E4B (Reasoning) | gpt-oss-20b (high) |
|---|---|---|
| Remote sensing and raster analysis | 1 | 11 |
| GIS analysis | 3 | 12 |
| Change detection | 8 | 12 |
| Climate and time series | 6 | 4 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 · gpt-oss-20b (high), Codex.
Side by side
| Metric | Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 | gpt-oss-20b (high), Codex |
|---|---|---|
| Geospatial Agent Index: Text only | 4 | 9 |
| Geospatial Agent Index: Text + vision | No data | No image input |
| Earth-Bench V2 score | 5 | 6 |
| GeoBenchX score | 3 | 10 |
| GS-QA2 score | 3 | 12 |
| Vision agents score | No data | No image input |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 4 | 10 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 5 | 11 |
| OGD4All score (standalone; in no index) | 21 | No data |
| GeoNatureAgent score (standalone; in no index) | 0 | No data |
| MultiGlobeQA score (standalone; in no index) | 16 | No data |
| GeoAgentEval score (standalone; in no index) | 40 | No data |
| ImpactMesh score (standalone; in no index) | No data | No data |
| Remote sensing and raster analysis score | 1 | 11 |
| Change detection score | 8 | 12 |
| GIS analysis score | 3 | 12 |
| Climate and time series score | 6 | 4 |
| Making maps score | 2 | 4 |
| Cost per task | No data | $0.094 |
| Time per task (at 100 output tokens/s) | 5.2 min | 4.8 min |
| Tokens per attempt | 341k | 457k |
| Turns per attempt | 24.4 | 24.2 |
| Input price per 1M tokens | No data | $0.070 |
| Output price per 1M tokens | No data | $0.30 |
| Context window | 65,536 | 128,000 |
| Image input | No | No |
| Developer | NVIDIA | OpenAI |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 or gpt-oss-20b (high), Codex?
gpt-oss-20b (high), Codex, 9 to 4.
Which is cheaper per task, Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 or gpt-oss-20b (high), Codex?
Not known yet: Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 has no cost data.
Which writes fewer output tokens per task, Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 or gpt-oss-20b (high), Codex?
Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2, 14k to 23k.
Which has the larger context window, Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2 or gpt-oss-20b (high), Codex?
gpt-oss-20b (high), Codex, 128,000 tokens to 65,536 tokens.