Llama 3.3 70B (Non-reasoning) vs Nemotron 3.5 Super 120B (Reasoning)
Results updated 9 November 2026 · Both in Terminus-2 · How we benchmark
Verdict
Nemotron 3.5 Super 120B (Reasoning) clearly beats Llama 3.3 70B (Non-reasoning) by 48 points, and is cheaper; Llama 3.3 70B (Non-reasoning) is faster.
- Cost per task
- $0.023 vs $0.011
- Time per task
- 2.1 min vs 13.8 min
- Tokens per attempt
- 62k vs 105k
- Image input
- No vs No
Nemotron 3.5 Super 120B (Reasoning), Terminus-2 scores higher than Llama 3.3 70B (Non-reasoning), Terminus-2 on all 3 evaluations scored for both. Nemotron 3.5 Super 120B (Reasoning), Terminus-2 is cheaper ($0.011 against $0.023 per task), but Llama 3.3 70B (Non-reasoning), Terminus-2 is faster (2.1 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 | Llama 3.3 70B (Non-reasoning) | Nemotron 3.5 Super 120B (Reasoning) |
|---|---|---|
| Making maps | 2 | 70 |
| GIS analysis | 12 | 70 |
| Change detection | 4 | 56 |
| Remote sensing and raster analysis | 7 | 54 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Llama 3.3 70B (Non-reasoning), Terminus-2 · Nemotron 3.5 Super 120B (Reasoning), Terminus-2.
Side by side
| Metric | Llama 3.3 70B (Non-reasoning), Terminus-2 | Nemotron 3.5 Super 120B (Reasoning), Terminus-2 |
|---|---|---|
| Geospatial Agent Index: Text only | 8 | 56 |
| Geospatial Agent Index: Text + vision | No image input | No data |
| Earth-Bench V2 score | 9 | 49 |
| GeoBenchX score | 6 | 56 |
| GS-QA2 score | 10 | 63 |
| Vision agents score | No image input | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 8 | 83 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 0 | 79 |
| OGD4All score (standalone; in no index) | No data | 85 |
| GeoNatureAgent score (standalone; in no index) | No data | 100 |
| MultiGlobeQA score (standalone; in no index) | No data | 70 |
| GeoAgentEval score (standalone; in no index) | No data | 84 |
| ImpactMesh score (standalone; in no index) | No data | No data |
| Remote sensing and raster analysis score | 7 | 54 |
| Change detection score | 4 | 56 |
| GIS analysis score | 12 | 70 |
| Climate and time series score | 12 | 50 |
| Making maps score | 2 | 70 |
| Cost per task | $0.023 | $0.011 |
| Time per task (at 100 output tokens/s) | 2.1 min | 13.8 min |
| Tokens per attempt | 62k | 105k |
| Turns per attempt | 14.4 | 10.5 |
| Input price per 1M tokens | $0.37 | $0.20 |
| Output price per 1M tokens | $0.56 | $0.50 |
| Context window | 24,000 | 262,144 |
| Image input | No | No |
| Developer | Meta | NVIDIA |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Llama 3.3 70B (Non-reasoning), Terminus-2 or Nemotron 3.5 Super 120B (Reasoning), Terminus-2?
Nemotron 3.5 Super 120B (Reasoning), Terminus-2, 56 to 8.
Which is cheaper per task, Llama 3.3 70B (Non-reasoning), Terminus-2 or Nemotron 3.5 Super 120B (Reasoning), Terminus-2?
Nemotron 3.5 Super 120B (Reasoning), Terminus-2, $0.011 to $0.023.
Which writes fewer output tokens per task, Llama 3.3 70B (Non-reasoning), Terminus-2 or Nemotron 3.5 Super 120B (Reasoning), Terminus-2?
Llama 3.3 70B (Non-reasoning), Terminus-2, 2k to 8k.
Which has the larger context window, Llama 3.3 70B (Non-reasoning), Terminus-2 or Nemotron 3.5 Super 120B (Reasoning), Terminus-2?
Nemotron 3.5 Super 120B (Reasoning), Terminus-2, 262,144 tokens to 24,000 tokens.