Gemma 4 31B (Reasoning) vs GPT-5.6 Luna (low)
Results updated 11 November 2026 · In Terminus-2 and Codex, so part of any gap can come from the harness · How we benchmark
Verdict
GPT-5.6 Luna (low) beats Gemma 4 31B (Reasoning) by 6 points, and is cheaper and faster.
- Cost per task
- $0.015 vs $0.0058
- Time per task
- 2.1 min vs 0.5 min
- Tokens per attempt
- 93k vs 84k
- Image input
- Yes vs Yes
GPT-5.6 Luna (low), Codex scores higher on GeoBenchX and GS-QA2; they are level on Earth-Bench V2. GPT-5.6 Luna (low), Codex is cheaper and faster: $0.0058 against $0.015 per task, and 0.5 min against 2.1 min.
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) | GPT-5.6 Luna (low) |
|---|---|---|
| GIS analysis | 56 | 72 |
| Remote sensing and raster analysis | 59 | 72 |
| Making maps | 70 | 81 |
| Climate and time series | 47 | 40 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Gemma 4 31B (Reasoning), Terminus-2 · GPT-5.6 Luna (low), Codex.
Side by side
| Metric | Gemma 4 31B (Reasoning), Terminus-2 | GPT-5.6 Luna (low), Codex |
|---|---|---|
| Geospatial Agent Index: Text only | 55 | 61 |
| Geospatial Agent Index: Text + vision | No data | 62 |
| Earth-Bench V2 score | 44 | 44 |
| GeoBenchX score | 61 | 62 |
| GS-QA2 score | 61 | 76 |
| Vision agents score | No data | 64 |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 83 | 86 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 63 | 95 |
| 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 | 72 |
| Change detection score | 52 | 52 |
| GIS analysis score | 56 | 72 |
| Climate and time series score | 47 | 40 |
| Making maps score | 70 | 81 |
| Cost per task | $0.015 | $0.0058 |
| Time per task (at 100 output tokens/s) | 2.1 min | 0.5 min |
| Tokens per attempt | 93k | 84k |
| Turns per attempt | 9.0 | 7.0 |
| Input price per 1M tokens | $0.17 | $0.20 |
| Output price per 1M tokens | $0.40 | $1.20 |
| Context window | 262,144 | 1,050,000 |
| Image input | Yes | Yes |
| Developer | OpenAI |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Gemma 4 31B (Reasoning), Terminus-2 or GPT-5.6 Luna (low), Codex?
GPT-5.6 Luna (low), Codex, 61 to 55.
Which is cheaper per task, Gemma 4 31B (Reasoning), Terminus-2 or GPT-5.6 Luna (low), Codex?
GPT-5.6 Luna (low), Codex, $0.0058 to $0.015.
Which writes fewer output tokens per task, Gemma 4 31B (Reasoning), Terminus-2 or GPT-5.6 Luna (low), Codex?
GPT-5.6 Luna (low), Codex, 2k to 8k.
Which has the larger context window, Gemma 4 31B (Reasoning), Terminus-2 or GPT-5.6 Luna (low), Codex?
GPT-5.6 Luna (low), Codex, 1,050,000 tokens to 262,144 tokens.