Terminus-2 – Gemma 4 26B A4B (thinking on) vs. Codex – GPT-6 Luna (low): geospatial agent comparison

Comparison between Terminus-2 – Gemma 4 26B A4B (thinking on) and Codex – GPT-6 Luna (low) on the Geospatial Agent Index, each benchmark, cost, time and token use, from the same tasks, agent and limits. Methodology.

Side by side

Terminus-2 – Gemma 4 26B A4B (thinking on) vs. Codex – GPT-6 Luna (low)
Metric	Terminus-2 – Gemma 4 26B A4B (thinking on)	Codex – GPT-6 Luna (low)
Geospatial Agent Index	58* (partial: attempt 1 on 457 of 488 tasks)	51* (partial: attempt 1 on 449 of 488 tasks)
Earth-Bench V2 score	46	44
GeoBenchX score	63	50
GS-QA2 score	64	59
GeoAgentBench score (guided GIS workflow; outside the index)	86	92
GeoAnalystBench score (guided GIS workflow; outside the index)	63	79
Raster and remote sensing score	64	55
Vector and overlay score	60	57
Climate and time series score	47	49
Spatial statistics and interpolation score	31	38
Mapping and cartography score	70	71
Recognising infeasible requests score	63	30
Cost per task	$0.018	$0.0031
Time per task	11.5 min	1.4 min
Tokens per attempt	142k	114k
Turns per attempt	13.2	7.8
Attempts decided	457	449
Input price per 1M tokens	$0.10	$0.10
Output price per 1M tokens	$0.30	$0.50
Context window	256,000	No data
Image input	Yes	No data
Developer	Google	OpenAI

Charts

Partial coverage. Terminus-2 – Gemma 4 26B A4B (thinking on): 190 of 221 GS-QA2 tasks; Codex – GPT-6 Luna (low): 111 of 150 GeoBenchX tasks. Grey placeholders identify these entries in the index, cost and output token charts; their height does not represent a value. Comparable bars appear when first-attempt coverage is complete. Their available results remain in the side-by-side table.

Geospatial Agent Index

Equal-weight average of the evaluation scores, 0 to 100 · Higher is better

Data table: Geospatial Agent Index

Geospatial Agent Index
Model	Creator	Geospatial Agent Index	Range while attempts are pending	Benchmarks covered	Coverage (attempts)

Cost per task

Average cost per task (USD) · Lower is better

How to read this chart

Not shown: partial coverage. 2 entries without a first attempt on every index task. What is still running is on the models page.

Data table: Cost per task

Cost per task
Model	Creator	Cost per task	Coverage (attempts)

Output tokens per task

Average output tokens per attempt, reasoning included · Lower is better

How to read this chart

Not shown: partial coverage. 2 entries without a first attempt on every index task. What is still running is on the models page.

Data table: Output tokens per task

Output tokens per task
Model	Creator	Output tokens per task	Coverage (attempts)

Frequently asked questions

Which scores higher on the Geospatial Agent Index, Terminus-2 – Gemma 4 26B A4B (thinking on) or Codex – GPT-6 Luna (low)?

Not comparable yet: Terminus-2 – Gemma 4 26B A4B (thinking on) has partial coverage (190 of 221 GS-QA2 tasks) and Codex – GPT-6 Luna (low) has partial coverage (111 of 150 GeoBenchX tasks), so it is not ranked. Terminus-2 – Gemma 4 26B A4B (thinking on): 457 of 722 scheduled attempts. Codex – GPT-6 Luna (low): 449 of 722 scheduled attempts. Coverage is partial; scores can change as more attempts are decided.

Which is cheaper per task, Terminus-2 – Gemma 4 26B A4B (thinking on) or Codex – GPT-6 Luna (low)?

Codex – GPT-6 Luna (low), at $0.0031 against $0.018 for Terminus-2 – Gemma 4 26B A4B (thinking on).

Which writes fewer output tokens per task, Terminus-2 – Gemma 4 26B A4B (thinking on) or Codex – GPT-6 Luna (low)?

Codex – GPT-6 Luna (low), at 1k against 18k for Terminus-2 – Gemma 4 26B A4B (thinking on).

Which has the larger context window, Terminus-2 – Gemma 4 26B A4B (thinking on) or Codex – GPT-6 Luna (low)?

Not known yet: Codex – GPT-6 Luna (low) has no context window data.

More: Terminus-2 – Gemma 4 26B A4B (thinking on) · Codex – GPT-6 Luna (low) · all comparisons