Gemma 4 31B (Reasoning) vs Llama 3.3 70B (Non-reasoning)
Results updated 11 November 2026 · Both in Terminus-2 · How we benchmark
Verdict
Gemma 4 31B (Reasoning) clearly beats Llama 3.3 70B (Non-reasoning) by 47 points, and is cheaper.
- Cost per task
- $0.015 vs $0.023
- Time per task
- 2.1 min vs 2.1 min
- Tokens per attempt
- 93k vs 62k
- Image input
- Yes vs No
Gemma 4 31B (Reasoning), Terminus-2 scores higher than Llama 3.3 70B (Non-reasoning), Terminus-2 on all 3 evaluations scored for both. Gemma 4 31B (Reasoning), Terminus-2 is cheaper: $0.015 against $0.023 per task.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Gemma 4 31B (Reasoning) | Llama 3.3 70B (Non-reasoning) |
|---|---|---|
| Making maps | 70 | 2 |
| Remote sensing and raster analysis | 59 | 7 |
| Change detection | 52 | 4 |
| GIS analysis | 56 | 12 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Gemma 4 31B (Reasoning), Terminus-2 · Llama 3.3 70B (Non-reasoning), Terminus-2.
Side by side
| Metric | Gemma 4 31B (Reasoning), Terminus-2 | Llama 3.3 70B (Non-reasoning), Terminus-2 |
|---|---|---|
| Geospatial Agent Index: Text only | 55 | 8 |
| Geospatial Agent Index: Text + vision | No data | No image input |
| Earth-Bench V2 score | 44 | 9 |
| GeoBenchX score | 61 | 6 |
| GS-QA2 score | 61 | 10 |
| Vision agents score | No data | No image input |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 83 | 8 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 63 | 0 |
| OGD4All score (standalone; in no index) | 85 | No data |
| GeoNatureAgent score (standalone; in no index) | 100 | No data |
| MultiGlobeQA score (standalone; in no index) | 30 | No data |
| GeoAgentEval score (standalone; in no index) | 88 | No data |
| ImpactMesh score (standalone; in no index) | No data | No data |
| Remote sensing and raster analysis score | 59 | 7 |
| Change detection score | 52 | 4 |
| GIS analysis score | 56 | 12 |
| Climate and time series score | 47 | 12 |
| Making maps score | 70 | 2 |
| Cost per task | $0.015 | $0.023 |
| Time per task (at 100 output tokens/s) | 2.1 min | 2.1 min |
| Tokens per attempt | 93k | 62k |
| Turns per attempt | 9.0 | 14.4 |
| Input price per 1M tokens | $0.17 | $0.37 |
| Output price per 1M tokens | $0.40 | $0.56 |
| Context window | 262,144 | 24,000 |
| Image input | Yes | No |
| Developer | Meta |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Gemma 4 31B (Reasoning), Terminus-2 or Llama 3.3 70B (Non-reasoning), Terminus-2?
Gemma 4 31B (Reasoning), Terminus-2, 55 to 8.
Which is cheaper per task, Gemma 4 31B (Reasoning), Terminus-2 or Llama 3.3 70B (Non-reasoning), Terminus-2?
Gemma 4 31B (Reasoning), Terminus-2, $0.015 to $0.023.
Which writes fewer output tokens per task, Gemma 4 31B (Reasoning), Terminus-2 or Llama 3.3 70B (Non-reasoning), Terminus-2?
Llama 3.3 70B (Non-reasoning), Terminus-2, 2k to 8k.
Which has the larger context window, Gemma 4 31B (Reasoning), Terminus-2 or Llama 3.3 70B (Non-reasoning), Terminus-2?
Gemma 4 31B (Reasoning), Terminus-2, 262,144 tokens to 24,000 tokens.