Geospatial Intelligence Index
Which AI model is best at geospatial (GIS and Earth observation) work: independent tests of agent tasks, vision, guided workflows and geospatial knowledge, scored the same way for every model.
New: Version 1.0: the first public release, with the same method as Pre-release 0.34.
Ongoing evaluation: results update as runs complete. Last updated 23 October 2026.
Headline results
Intelligence
Geospatial Intelligence Index, Text + vision, 0 to 100 · Higher is better
Top eight. Every model is in the table.
How to read this chart
Rows. One bar per model, at its highest reasoning level that is ranked. The level is in brackets. Other levels are in the leaderboard.
What this metric means. A weighted average of four group scores, each 0 to 100: Agent tasks 4/8, Vision 2/8, Guided workflows 1/8, Geospatial knowledge 1/8. Agent tasks carry half because they test the work itself. Agent tasks here means the Geospatial Agent Index, Text + vision view.
Not shown: not yet ranked. 13 models are not yet complete in every group. They are listed under "Not yet ranked" in the leaderboard, with what each still needs.
Data table: Intelligence
| Rank | Model | Developer | Index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 82.7 | 88 | 59 | 99 | 94 |
| 2 | Claude Sonnet 5.5 (xhigh) | Anthropic | 78.9 | 82 | 57 | 96 | 94 |
| 3 | GPT-6.1 Sol (high) (Highest ranked level; GPT-6.1 Sol (xhigh) is not yet ranked) | OpenAI | 78.1 | 82 | 53 | 100 | 93 |
| 4 | GPT-6 Sol (high) | OpenAI | 77.8 | 80 | 56 | 100 | 91 |
| 5 | Claude Opus 5 (high) | Anthropic | 77.4 | 81 | 53 | 95 | 93 |
| 6 | GPT-5.6 Sol (high) | OpenAI | 75.5 | 75 | 58 | 95 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 75.1 | 77 | 52 | 97 | 92 |
| 8 | Claude Opus 4.8 (high) | Anthropic | 74.0 | 75 | 50 | 98 | 93 |
Speed
Time per task (at 100 output tokens/s), minutes, models ranked in Text only · Lower is better
Top eight. Every model is in the table.
How to read this chart
Time per task. Time as if every model wrote 100 output tokens per second: output tokens divided by 100, plus the time the agent's commands took. Serving speed differs between providers and is not ranked. The average over the Geospatial Agent Index tasks, in the harness the model is scored in.
Data table: Speed
| Model | Developer | Time per task (min) | Tasks attempted |
|---|---|---|---|
| gpt-oss-120b (low) | OpenAI | 0.5 min | 488 of 488 tasks |
| GPT-6 Sol (high) | OpenAI | 0.8 min | 488 of 488 tasks |
| GPT-6.1 Sol (xhigh) | OpenAI | 0.8 min | 488 of 488 tasks |
| GPT-5.6 Terra (high) | OpenAI | 0.8 min | 488 of 488 tasks |
| Claude Opus 4.8 (high) | Anthropic | 0.9 min | 488 of 488 tasks |
| Claude Opus 4.7 (high) | Anthropic | 0.9 min | 488 of 488 tasks |
| GPT-5.6 Sol (high) | OpenAI | 1.0 min | 488 of 488 tasks |
| Claude Opus 5 (high) | Anthropic | 1.1 min | 488 of 488 tasks |
| Mistral Large 3 (Non-reasoning) | Mistral AI | 1.1 min | 488 of 488 tasks |
| Claude Sonnet 5.5 (xhigh) | Anthropic | 1.2 min | 488 of 488 tasks |
| Claude Opus 5.5 (xhigh) | Anthropic | 1.4 min | 488 of 488 tasks |
| GPT-6 Luna (xhigh) | OpenAI | 1.4 min | 488 of 488 tasks |
| Claude Sonnet 5 (high) | Anthropic | 1.5 min | 488 of 488 tasks |
| Claude Haiku 4.5 (Reasoning) | Anthropic | 1.6 min | 488 of 488 tasks |
| GPT-5.6 Luna (xhigh) | OpenAI | 1.7 min | 488 of 488 tasks |
| Llama 3.3 70B (Non-reasoning) | Meta | 2.1 min | 488 of 488 tasks |
| Claude Haiku 5.5 (xhigh) | Anthropic | 2.5 min | 488 of 488 tasks |
| Kimi K3 (Reasoning) | Moonshot AI | 2.7 min | 488 of 488 tasks |
| GLM-5.3 (max) | Z.ai | 4.2 min | 488 of 488 tasks |
| DeepSeek V4 Pro (max) | DeepSeek | 4.9 min | 488 of 488 tasks |
| GLM-4.7-Flash (Reasoning) | Z.ai | 5.4 min | 488 of 488 tasks |
| Gemma 4 26B A4B (Reasoning) | 5.6 min | 488 of 488 tasks | |
| DeepSeek V4 Flash (max) | DeepSeek | 5.7 min | 488 of 488 tasks |
| GLM-5.2 (max) | Z.ai | 6.1 min | 488 of 488 tasks |
| GLM-5.3-Flash (max) | Z.ai | 8.5 min | 488 of 488 tasks |
Cost per task
Average cost of one agent task (USD), models ranked in Text only · Lower is better
Top eight. Every model is in the table.
How to read this chart
Cost per agent task. The average cost of one Geospatial Agent Index task, at public API list prices, in the harness the model is scored in. Knowledge and image questions are single model calls and are not included.
Data table: Cost per task
| Model | Developer | Cost per task | Tasks attempted |
|---|---|---|---|
| gpt-oss-120b (low) | OpenAI | $0.0046 | 488 of 488 tasks |
| GPT-6 Luna (xhigh) | OpenAI | $0.0070 | 488 of 488 tasks |
| Claude Haiku 5.5 (xhigh) | Anthropic | $0.017 | 488 of 488 tasks |
| GPT-5.6 Luna (xhigh) | OpenAI | $0.021 | 488 of 488 tasks |
| Gemma 4 26B A4B (Reasoning) | $0.021 | 488 of 488 tasks | |
| Llama 3.3 70B (Non-reasoning) | Meta | $0.023 | 488 of 488 tasks |
| Mistral Large 3 (Non-reasoning) | Mistral AI | $0.026 | 488 of 488 tasks |
| GLM-4.7-Flash (Reasoning) | Z.ai | $0.037 | 488 of 488 tasks |
| DeepSeek V4 Flash (max) | DeepSeek | $0.062 | 488 of 488 tasks |
| GPT-6 Sol (high) | OpenAI | $0.082 | 488 of 488 tasks |
| GPT-6.1 Sol (xhigh) | OpenAI | $0.085 | 488 of 488 tasks |
| Claude Haiku 4.5 (Reasoning) | Anthropic | $0.092 | 488 of 488 tasks |
| GLM-5.3-Flash (max) | Z.ai | $0.094 | 488 of 488 tasks |
| GPT-5.6 Terra (high) | OpenAI | $0.10 | 488 of 488 tasks |
| Claude Sonnet 5.5 (xhigh) | Anthropic | $0.11 | 488 of 488 tasks |
| Claude Sonnet 5 (high) | Anthropic | $0.20 | 488 of 488 tasks |
| GPT-5.6 Sol (high) | OpenAI | $0.21 | 488 of 488 tasks |
| Kimi K3 (Reasoning) | Moonshot AI | $0.23 | 488 of 488 tasks |
| Claude Opus 5.5 (xhigh) | Anthropic | $0.25 | 488 of 488 tasks |
| Claude Opus 4.8 (high) | Anthropic | $0.26 | 488 of 488 tasks |
| Claude Opus 5 (high) | Anthropic | $0.32 | 488 of 488 tasks |
| GLM-5.2 (max) | Z.ai | $0.32 | 488 of 488 tasks |
| Claude Opus 4.7 (high) | Anthropic | $0.34 | 488 of 488 tasks |
| DeepSeek V4 Pro (max) | DeepSeek | $0.35 | 488 of 488 tasks |
| GLM-5.3 (max) | Z.ai | $0.38 | 488 of 488 tasks |
Main index charts
Geospatial Intelligence Index
Text + vision adds tasks where the model must look at images. In Text only, models still work with satellite data through code.
Geospatial Intelligence Index: Text only
Weighted average of agent tasks, guided workflows and geospatial knowledge, 0 to 100 · Higher is better
How to read this chart
Rows. One bar per model, at its highest reasoning level that is ranked. The level is in brackets. Other levels are in the leaderboard.
What this metric means. A weighted average of three group scores, each 0 to 100: Agent tasks 4/6, Guided workflows 1/6, Geospatial knowledge 1/6. Agent tasks here means the Geospatial Agent Index, Text only view (Earth-Bench V2, GeoBenchX and GS-QA2). Every model complete in these groups is ranked, models with no image input included.
Not shown: not yet ranked. 9 models are not yet complete in every group of this view. They are listed under "Not yet ranked" in the leaderboard, with what each still needs.
Data table: Geospatial Intelligence Index: Text only
| Rank | Model | Developer | Index | Agent tasks (4/6) | Guided workflows (1/6) | Geospatial knowledge (1/6) |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 88.7 | 85 | 99 | 94 |
| 2 | GPT-6.1 Sol (xhigh) | OpenAI | 86.2 | 82 | 97 | 93 |
| 3 | GPT-6 Sol (high) | OpenAI | 84.9 | 80 | 100 | 91 |
| 4 | Claude Opus 5 (high) | Anthropic | 84.5 | 80 | 95 | 93 |
| 5 | Claude Sonnet 5.5 (xhigh) | Anthropic | 84.4 | 79 | 96 | 94 |
| 6 | Claude Opus 4.8 (high) | Anthropic | 82.5 | 76 | 98 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 82.0 | 76 | 97 | 92 |
| 8 | GPT-5.6 Sol (high) | OpenAI | 81.3 | 75 | 95 | 93 |
| 9 | Claude Sonnet 5 (high) | Anthropic | 80.9 | 74 | 96 | 91 |
| 10 | Claude Opus 4.7 (high) | Anthropic | 80.8 | 74 | 98 | 93 |
| 11 | GPT-5.6 Luna (xhigh) | OpenAI | 80.0 | 73 | 99 | 91 |
| 12 | GPT-6 Luna (xhigh) | OpenAI | 80.0 | 73 | 98 | 91 |
| 13 | DeepSeek V4 Pro (max) Open weights (No image input) | DeepSeek | 79.0 | 71 | 98 | 92 |
| 14 | Kimi K3 (Reasoning) Open weights | Moonshot AI | 78.6 | 72 | 92 | 92 |
| 15 | GLM-5.2 (max) Open weights (No image input) | Z.ai | 78.4 | 72 | 94 | 89 |
| 16 | GPT-5.6 Terra (high) | OpenAI | 77.5 | 69 | 95 | 92 |
| 17 | DeepSeek V4 Flash (max) Open weights (No image input) | DeepSeek | 77.2 | 69 | 95 | 91 |
| 18 | GLM-5.3 (max) Open weights (No image input) | Z.ai | 77.1 | 70 | 89 | 92 |
| 19 | GLM-5.3-Flash (max) Open weights | Z.ai | 71.0 | 63 | 84 | 92 |
| 20 | Claude Haiku 4.5 (Reasoning) | Anthropic | 65.8 | 58 | 76 | 89 |
| 21 | Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | 56 | 75 | 89 | |
| 22 | GLM-4.7-Flash (Reasoning) Open weights (No image input) | Z.ai | 49.6 | 43 | 39 | 86 |
| 23 | Mistral Large 3 (Non-reasoning) Open weights | Mistral AI | 44.5 | 33 | 47 | 89 |
| 24 | gpt-oss-120b (low) Open weights (Highest ranked level; gpt-oss-120b (high) is not yet ranked) (No image input) | OpenAI | 44.3 | 31 | 54 | 88 |
| 25 | Llama 3.3 70B (Non-reasoning) Open weights (No image input) | Meta | 19.9 | 8 | 4 | 82 |
Geospatial Intelligence Index: Text + vision
Weighted average of agent tasks, vision, guided workflows and geospatial knowledge, 0 to 100 · Higher is better
No image input (see Text only): DeepSeek V4 Pro (max) Open weights, GLM-5.2 (max) Open weights, DeepSeek V4 Flash (max) Open weights, GLM-5.3 (max) Open weights, GLM-4.7-Flash (Reasoning) Open weights, gpt-oss-120b (low) Open weights, Llama 3.3 70B (Non-reasoning) Open weights.
How to read this chart
Rows. One bar per model, at its highest reasoning level that is ranked. The level is in brackets. Other levels are in the leaderboard.
What this metric means. A weighted average of four group scores, each 0 to 100: Agent tasks 4/8, Vision 2/8, Guided workflows 1/8, Geospatial knowledge 1/8. Agent tasks carry half because they test the work itself. Agent tasks here means the Geospatial Agent Index, Text + vision view.
No image input. No image input: the developer's documentation says the model cannot take images. It is ranked in the Text only view and never given 0. Such models are listed under the chart.
Not shown: not yet ranked. 13 models are not yet complete in every group. They are listed under "Not yet ranked" in the leaderboard, with what each still needs.
Data table: Geospatial Intelligence Index: Text + vision
| Rank | Model | Developer | Index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 82.7 | 88 | 59 | 99 | 94 |
| 2 | Claude Sonnet 5.5 (xhigh) | Anthropic | 78.9 | 82 | 57 | 96 | 94 |
| 3 | GPT-6.1 Sol (high) (Highest ranked level; GPT-6.1 Sol (xhigh) is not yet ranked) | OpenAI | 78.1 | 82 | 53 | 100 | 93 |
| 4 | GPT-6 Sol (high) | OpenAI | 77.8 | 80 | 56 | 100 | 91 |
| 5 | Claude Opus 5 (high) | Anthropic | 77.4 | 81 | 53 | 95 | 93 |
| 6 | GPT-5.6 Sol (high) | OpenAI | 75.5 | 75 | 58 | 95 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 75.1 | 77 | 52 | 97 | 92 |
| 8 | Claude Opus 4.8 (high) | Anthropic | 74.0 | 75 | 50 | 98 | 93 |
| 9 | Claude Sonnet 5 (high) | Anthropic | 73.2 | 76 | 47 | 96 | 91 |
| 10 | Claude Opus 4.7 (high) | Anthropic | 72.8 | 74 | 47 | 98 | 93 |
| 11 | GPT-5.6 Luna (xhigh) | OpenAI | 72.8 | 72 | 52 | 99 | 91 |
| 12 | GPT-6 Luna (xhigh) | OpenAI | 72.3 | 72 | 50 | 98 | 91 |
| 13 | GPT-5.6 Terra (high) | OpenAI | 71.5 | 69 | 54 | 95 | 92 |
| 14 | Claude Haiku 4.5 (Reasoning) | Anthropic | 53.1 | 48 | 34 | 76 | 89 |
The top model leads the next by about 4 points; differences under about 3 points are within noise (how this is estimated).
Performance and cost
Text only
Geospatial Intelligence Index: Text only vs. cost per agent task
Geospatial Intelligence Index; cost per Agent Index task (USD) · Higher and further left is better
- Most attractive quadrant
- Pareto line
How to read this chart
How to read this chart. Each point is one model at its highest reasoning level. Further left means a lower average cost per agent task (log scale); higher means a higher Geospatial Intelligence Index. Both axes fit the models shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot, split at the middle of each axis: a reading aid, not a fixed bar. The dotted Pareto line joins the models that no cheaper model beats on the index.
Cost per agent task. The average cost of one Geospatial Agent Index task, at public API list prices, in the harness the model is scored in. Knowledge and image questions are single model calls and are not included.
Data table: Geospatial Intelligence Index: Text only vs. cost per agent task
| Model | Geospatial Intelligence Index | Cost per agent task (USD) | On the Pareto line |
|---|---|---|---|
| Claude Opus 5.5 (xhigh) | 88.7 | $0.25 | Yes |
| GPT-6.1 Sol (xhigh) | 86.2 | $0.085 | Yes |
| GPT-6 Sol (high) | 84.9 | $0.082 | Yes |
| Claude Opus 5 (high) | 84.5 | $0.32 | No |
| Claude Sonnet 5.5 (xhigh) | 84.4 | $0.11 | No |
| Claude Opus 4.8 (high) | 82.5 | $0.26 | No |
| Claude Haiku 5.5 (xhigh) | 82.0 | $0.017 | Yes |
| GPT-5.6 Sol (high) | 81.3 | $0.21 | No |
| Claude Sonnet 5 (high) | 80.9 | $0.20 | No |
| Claude Opus 4.7 (high) | 80.8 | $0.34 | No |
| GPT-5.6 Luna (xhigh) | 80.0 | $0.021 | No |
| GPT-6 Luna (xhigh) | 80.0 | $0.0070 | Yes |
| DeepSeek V4 Pro (max) Open weights | 79.0 | $0.35 | No |
| Kimi K3 (Reasoning) Open weights | 78.6 | $0.23 | No |
| GLM-5.2 (max) Open weights | 78.4 | $0.32 | No |
| GPT-5.6 Terra (high) | 77.5 | $0.10 | No |
| DeepSeek V4 Flash (max) Open weights | 77.2 | $0.062 | No |
| GLM-5.3 (max) Open weights | 77.1 | $0.38 | No |
| GLM-5.3-Flash (max) Open weights | 71.0 | $0.094 | No |
| Claude Haiku 4.5 (Reasoning) | 65.8 | $0.092 | No |
| Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | $0.021 | No |
| GLM-4.7-Flash (Reasoning) Open weights | 49.6 | $0.037 | No |
| Mistral Large 3 (Non-reasoning) Open weights | 44.5 | $0.026 | No |
| gpt-oss-120b (low) Open weights | 44.3 | $0.0046 | Yes |
| Llama 3.3 70B (Non-reasoning) Open weights | 19.9 | $0.023 | No |
Text + vision
Geospatial Intelligence Index: Text + vision vs. cost per agent task
Geospatial Intelligence Index; cost per Agent Index task (USD) · Higher and further left is better
- Most attractive quadrant
- Pareto line
How to read this chart
How to read this chart. Each point is one model at its highest reasoning level. Further left means a lower average cost per agent task (log scale); higher means a higher Geospatial Intelligence Index. Both axes fit the models shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot, split at the middle of each axis: a reading aid, not a fixed bar. The dotted Pareto line joins the models that no cheaper model beats on the index.
Cost per agent task. The average cost of one Geospatial Agent Index task, at public API list prices, in the harness the model is scored in. Knowledge and image questions are single model calls and are not included.
Data table: Geospatial Intelligence Index: Text + vision vs. cost per agent task
| Model | Geospatial Intelligence Index | Cost per agent task (USD) | On the Pareto line |
|---|---|---|---|
| Claude Opus 5.5 (xhigh) | 82.7 | $0.25 | Yes |
| Claude Sonnet 5.5 (xhigh) | 78.9 | $0.11 | Yes |
| GPT-6.1 Sol (high) | 78.1 | $0.059 | Yes |
| GPT-6 Sol (high) | 77.8 | $0.082 | No |
| Claude Opus 5 (high) | 77.4 | $0.32 | No |
| GPT-5.6 Sol (high) | 75.5 | $0.21 | No |
| Claude Haiku 5.5 (xhigh) | 75.1 | $0.017 | Yes |
| Claude Opus 4.8 (high) | 74.0 | $0.26 | No |
| Claude Sonnet 5 (high) | 73.2 | $0.20 | No |
| Claude Opus 4.7 (high) | 72.8 | $0.34 | No |
| GPT-5.6 Luna (xhigh) | 72.8 | $0.021 | No |
| GPT-6 Luna (xhigh) | 72.3 | $0.0070 | Yes |
| GPT-5.6 Terra (high) | 71.5 | $0.10 | No |
| Claude Haiku 4.5 (Reasoning) | 53.1 | $0.092 | No |
Leaderboard
Every model, reasoning level and group score
Every group score is shown next to the index, 0 to 100.
Text only
Text only: agent tasks 4/6, guided workflows 1/6, geospatial knowledge 1/6. Agent tasks here is the Geospatial Agent Index on its text evaluations (Earth-Bench V2, GeoBenchX and GS-QA2). Every model is ranked here, models with no image input included.
| Rank | Model | Developer | Index | Agent tasks (4/6) | Guided workflows (1/6) | Geospatial knowledge (1/6) |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 88.7 | 85 | 99 | 94 |
| 2 | GPT-6.1 Sol (xhigh) | OpenAI | 86.2 | 82 | 97 | 93 |
| 3 | GPT-6 Sol (high) | OpenAI | 84.9 | 80 | 100 | 91 |
| 4 | Claude Opus 5 (high) | Anthropic | 84.5 | 80 | 95 | 93 |
| 5 | Claude Sonnet 5.5 (xhigh) | Anthropic | 84.4 | 79 | 96 | 94 |
| 6 | Claude Opus 4.8 (high) | Anthropic | 82.5 | 76 | 98 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 82.0 | 76 | 97 | 92 |
| 8 | GPT-5.6 Sol (high) | OpenAI | 81.3 | 75 | 95 | 93 |
| 9 | Claude Sonnet 5 (high) | Anthropic | 80.9 | 74 | 96 | 91 |
| 10 | Claude Opus 4.7 (high) | Anthropic | 80.8 | 74 | 98 | 93 |
| 11 | GPT-5.6 Luna (xhigh) | OpenAI | 80.0 | 73 | 99 | 91 |
| 12 | GPT-6 Luna (xhigh) | OpenAI | 80.0 | 73 | 98 | 91 |
| 13 | DeepSeek V4 Pro (max) Open weights (No image input) | DeepSeek | 79.0 | 71 | 98 | 92 |
| 14 | Kimi K3 (Reasoning) Open weights | Moonshot AI | 78.6 | 72 | 92 | 92 |
| 15 | GLM-5.2 (max) Open weights (No image input) | Z.ai | 78.4 | 72 | 94 | 89 |
| 16 | GPT-5.6 Terra (high) | OpenAI | 77.5 | 69 | 95 | 92 |
| 17 | DeepSeek V4 Flash (max) Open weights (No image input) | DeepSeek | 77.2 | 69 | 95 | 91 |
| 18 | GLM-5.3 (max) Open weights (No image input) | Z.ai | 77.1 | 70 | 89 | 92 |
| 19 | GLM-5.3-Flash (max) Open weights | Z.ai | 71.0 | 63 | 84 | 92 |
| 20 | Claude Haiku 4.5 (Reasoning) | Anthropic | 65.8 | 58 | 76 | 89 |
| 21 | Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | 56 | 75 | 89 | |
| 22 | GLM-4.7-Flash (Reasoning) Open weights (No image input) | Z.ai | 49.6 | 43 | 39 | 86 |
| 23 | Mistral Large 3 (Non-reasoning) Open weights | Mistral AI | 44.5 | 33 | 47 | 89 |
| 24 | gpt-oss-120b (low) Open weights (Highest ranked level; gpt-oss-120b (high) is not yet ranked) (No image input) | OpenAI | 44.3 | 31 | 54 | 88 |
| 25 | Llama 3.3 70B (Non-reasoning) Open weights (No image input) | Meta | 19.9 | 8 | 4 | 82 |
One bar per model, at its highest reasoning level that is ranked. The level is in brackets. Other levels are under 'Show every reasoning level'.
Show every reasoning level
| Rank | Model | Developer | Index | Agent tasks (4/6) | Guided workflows (1/6) | Geospatial knowledge (1/6) |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 88.7 | 85 | 99 | 94 |
| 2 | Claude Opus 5.5 (high) | Anthropic | 87.4 | 82 | 100 | 95 |
| 3 | GPT-6.1 Sol (medium) | OpenAI | 86.4 | 82 | 100 | 92 |
| 4 | GPT-6.1 Sol (xhigh) | OpenAI | 86.2 | 82 | 97 | 93 |
| 5 | GPT-6.1 Sol (high) | OpenAI | 85.7 | 80 | 100 | 93 |
| 6 | Claude Opus 5.5 (medium) | Anthropic | 85.4 | 80 | 96 | 94 |
| 7 | GPT-6 Sol (high) | OpenAI | 84.9 | 80 | 100 | 91 |
| 8 | Claude Opus 5 (high) | Anthropic | 84.5 | 80 | 95 | 93 |
| 9 | Claude Sonnet 5.5 (xhigh) | Anthropic | 84.4 | 79 | 96 | 94 |
| 10 | Claude Sonnet 5.5 (high) | Anthropic | 84.3 | 78 | 99 | 94 |
| 11 | Claude Opus 4.8 (high) | Anthropic | 82.5 | 76 | 98 | 93 |
| 12 | Claude Haiku 5.5 (xhigh) | Anthropic | 82.0 | 76 | 97 | 92 |
| 13 | GPT-5.6 Sol (high) | OpenAI | 81.3 | 75 | 95 | 93 |
| 14 | Claude Haiku 5.5 (high) | Anthropic | 81.0 | 76 | 91 | 92 |
| 15 | Claude Sonnet 5 (high) | Anthropic | 80.9 | 74 | 96 | 91 |
| 16 | Claude Opus 4.7 (high) | Anthropic | 80.8 | 74 | 98 | 93 |
| 17 | GPT-5.6 Luna (xhigh) | OpenAI | 80.0 | 73 | 99 | 91 |
| 18 | GPT-6 Luna (xhigh) | OpenAI | 80.0 | 73 | 98 | 91 |
| 19 | GPT-5.6 Luna (high) | OpenAI | 79.5 | 72 | 99 | 91 |
| 20 | Claude Haiku 5.5 (medium) | Anthropic | 79.2 | 72 | 94 | 92 |
| 21 | GPT-6 Luna (high) | OpenAI | 79.2 | 71 | 98 | 92 |
| 22 | DeepSeek V4 Pro (max) Open weights (No image input) | DeepSeek | 79.0 | 71 | 98 | 92 |
| 23 | Claude Sonnet 5.5 (medium) | Anthropic | 78.9 | 71 | 96 | 93 |
| 24 | GLM-5.2 (high) Open weights (No image input) | Z.ai | 78.6 | 71 | 97 | 91 |
| 25 | Kimi K3 (Reasoning) Open weights | Moonshot AI | 78.6 | 72 | 92 | 92 |
| 26 | DeepSeek V4 Pro (high) Open weights (No image input) | DeepSeek | 78.6 | 70 | 98 | 92 |
| 27 | DeepSeek V4 Flash (high) Open weights (No image input) | DeepSeek | 78.4 | 72 | 92 | 91 |
| 28 | GLM-5.2 (max) Open weights (No image input) | Z.ai | 78.4 | 72 | 94 | 89 |
| 29 | GPT-5.6 Terra (high) | OpenAI | 77.5 | 69 | 95 | 92 |
| 30 | DeepSeek V4 Flash (max) Open weights (No image input) | DeepSeek | 77.2 | 69 | 95 | 91 |
| 31 | GLM-5.3 (max) Open weights (No image input) | Z.ai | 77.1 | 70 | 89 | 92 |
| 32 | GPT-5.6 Luna (medium) | OpenAI | 75.9 | 66 | 99 | 91 |
| 33 | GLM-5.3 (low) Open weights (No image input) | Z.ai | 74.5 | 66 | 93 | 91 |
| 34 | GLM-5.3-Flash (low) Open weights | Z.ai | 71.1 | 62 | 94 | 85 |
| 35 | GLM-5.3-Flash (max) Open weights | Z.ai | 71.0 | 63 | 84 | 92 |
| 36 | GLM-5.2 (Non-reasoning) Open weights (No image input) | Z.ai | 70.9 | 63 | 83 | 89 |
| 37 | GPT-5.6 Luna (low) | OpenAI | 70.5 | 61 | 90 | 90 |
| 38 | GPT-6 Luna (medium) | OpenAI | 68.5 | 57 | 92 | 90 |
| 39 | Claude Haiku 4.5 (Reasoning) | Anthropic | 65.8 | 58 | 76 | 89 |
| 40 | Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | 56 | 75 | 89 | |
| 41 | GPT-6 Luna (low) | OpenAI | 63.2 | 51 | 85 | 89 |
| 42 | GLM-4.7-Flash (Reasoning) Open weights (No image input) | Z.ai | 49.6 | 43 | 39 | 86 |
| 43 | Mistral Large 3 (Non-reasoning) Open weights | Mistral AI | 44.5 | 33 | 47 | 89 |
| 44 | gpt-oss-120b (low) Open weights (No image input) | OpenAI | 44.3 | 31 | 54 | 88 |
| 45 | Llama 3.3 70B (Non-reasoning) Open weights (No image input) | Meta | 19.9 | 8 | 4 | 82 |
Text + vision
| Rank | Model | Developer | Index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 82.7 | 88 | 59 | 99 | 94 |
| 2 | Claude Sonnet 5.5 (xhigh) | Anthropic | 78.9 | 82 | 57 | 96 | 94 |
| 3 | GPT-6.1 Sol (high) (Highest ranked level; GPT-6.1 Sol (xhigh) is not yet ranked) | OpenAI | 78.1 | 82 | 53 | 100 | 93 |
| 4 | GPT-6 Sol (high) | OpenAI | 77.8 | 80 | 56 | 100 | 91 |
| 5 | Claude Opus 5 (high) | Anthropic | 77.4 | 81 | 53 | 95 | 93 |
| 6 | GPT-5.6 Sol (high) | OpenAI | 75.5 | 75 | 58 | 95 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 75.1 | 77 | 52 | 97 | 92 |
| 8 | Claude Opus 4.8 (high) | Anthropic | 74.0 | 75 | 50 | 98 | 93 |
| 9 | Claude Sonnet 5 (high) | Anthropic | 73.2 | 76 | 47 | 96 | 91 |
| 10 | Claude Opus 4.7 (high) | Anthropic | 72.8 | 74 | 47 | 98 | 93 |
| 11 | GPT-5.6 Luna (xhigh) | OpenAI | 72.8 | 72 | 52 | 99 | 91 |
| 12 | GPT-6 Luna (xhigh) | OpenAI | 72.3 | 72 | 50 | 98 | 91 |
| 13 | GPT-5.6 Terra (high) | OpenAI | 71.5 | 69 | 54 | 95 | 92 |
| 14 | Claude Haiku 4.5 (Reasoning) | Anthropic | 53.1 | 48 | 34 | 76 | 89 |
One bar per model, at its highest reasoning level that is ranked. The level is in brackets. Other levels are under 'Show every reasoning level'.
Not yet ranked
Ranked once every group is complete. Sorted by the Text only index.
| Model | Developer | Text only index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|
| Kimi K3 (Reasoning) Open weights | Moonshot AI | 78.6 | Not run: Vision agents (all 64 tasks) | 48 | 92 | 92 |
| GLM-5.3-Flash (max) Open weights | Z.ai | 71.0 | Not run: Vision agents (all 64 tasks) | Incomplete: 14 of 136 questions rejected by the provider | 84 | 92 |
| Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | Not run: Vision agents (all 64 tasks) | 43 | 75 | 89 | |
| Mistral Large 3 (Non-reasoning) Open weights | Mistral AI | 44.5 | Not run: Vision agents (all 64 tasks) | Not run | 47 | 89 |
| Claude Sonnet 4.6 (high) | Anthropic | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 98 | Not run: 845 of 1,685 questions |
| Gemini 3.1 Flash-Lite (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GeoBenchX (2 of 150 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | Not run: GeoAgentBench (2 of 50 tasks) | Not run | |
| Gemini 3.1 Pro (preview) (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 90 | Not run | |
| Gemini 3.8 Flash (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 53 | Not run | |
| Kimi K2.6 (Reasoning) Open weights | Moonshot AI | Not yet ranked | Not run: Vision agents (all 64 tasks) | 44 | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 92 |
| Kimi K2.7 Code (Reasoning) Open weights | Moonshot AI | Not yet ranked | Not run: Vision agents (all 64 tasks) | 44 | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 92 |
| Mistral Small 3.1 24B (Non-reasoning) Open weights | Mistral AI | Not yet ranked | No agent entry | 33 | No agent entry | 82 |
| Qwen 3.8 27B (xhigh) Open weights | Alibaba (Qwen) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GeoBenchX (all 150 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run: 8 of 136 questions | Not run: GeoAgentBench (2 of 50 tasks); GeoAnalystBench (14 of 19 tasks) | 91 |
| gpt-oss-20b (high) Open weights (No image input) | OpenAI | Not yet ranked | 9 | No image input | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 85 |
No image input (see Text only): DeepSeek V4 Pro (max) Open weights, GLM-5.2 (max) Open weights, DeepSeek V4 Flash (max) Open weights, GLM-5.3 (max) Open weights, GLM-4.7-Flash (Reasoning) Open weights, gpt-oss-120b (low) Open weights, Llama 3.3 70B (Non-reasoning) Open weights.
Show every reasoning level
| Rank | Model | Developer | Index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (xhigh) | Anthropic | 82.7 | 88 | 59 | 99 | 94 |
| 2 | Claude Sonnet 5.5 (xhigh) | Anthropic | 78.9 | 82 | 57 | 96 | 94 |
| 3 | GPT-6.1 Sol (high) | OpenAI | 78.1 | 82 | 53 | 100 | 93 |
| 4 | GPT-6 Sol (high) | OpenAI | 77.8 | 80 | 56 | 100 | 91 |
| 5 | Claude Opus 5 (high) | Anthropic | 77.4 | 81 | 53 | 95 | 93 |
| 6 | GPT-5.6 Sol (high) | OpenAI | 75.5 | 75 | 58 | 95 | 93 |
| 7 | Claude Haiku 5.5 (xhigh) | Anthropic | 75.1 | 77 | 52 | 97 | 92 |
| 8 | Claude Opus 4.8 (high) | Anthropic | 74.0 | 75 | 50 | 98 | 93 |
| 9 | Claude Sonnet 5 (high) | Anthropic | 73.2 | 76 | 47 | 96 | 91 |
| 10 | Claude Opus 4.7 (high) | Anthropic | 72.8 | 74 | 47 | 98 | 93 |
| 11 | GPT-5.6 Luna (xhigh) | OpenAI | 72.8 | 72 | 52 | 99 | 91 |
| 12 | GPT-6 Luna (xhigh) | OpenAI | 72.3 | 72 | 50 | 98 | 91 |
| 13 | GPT-5.6 Luna (high) | OpenAI | 72.1 | 71 | 51 | 99 | 91 |
| 14 | GPT-6 Luna (high) | OpenAI | 72.0 | 70 | 53 | 98 | 92 |
| 15 | GPT-5.6 Terra (high) | OpenAI | 71.5 | 69 | 54 | 95 | 92 |
| 16 | GPT-5.6 Luna (medium) | OpenAI | 69.1 | 67 | 47 | 99 | 91 |
| 17 | GPT-5.6 Luna (low) | OpenAI | 65.9 | 62 | 50 | 90 | 90 |
| 18 | GPT-6 Luna (medium) | OpenAI | 64.6 | 59 | 50 | 92 | 90 |
| 19 | GPT-6 Luna (low) | OpenAI | 56.4 | 45 | 48 | 85 | 89 |
| 20 | Claude Haiku 4.5 (Reasoning) | Anthropic | 53.1 | 48 | 34 | 76 | 89 |
Not yet ranked (every reasoning level)
Ranked once every group is complete. Sorted by the Text only index.
| Model | Developer | Text only index | Agent tasks (4/8) | Vision (2/8) | Guided workflows (1/8) | Geospatial knowledge (1/8) |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 (high) | Anthropic | 87.4 | Not run: Vision agents (all 64 tasks) | 59 | 100 | 95 |
| GPT-6.1 Sol (medium) | OpenAI | 86.4 | Not run: Vision agents (all 64 tasks) | 55 | 100 | 92 |
| GPT-6.1 Sol (xhigh) | OpenAI | 86.2 | Not run: Vision agents (4 of 64 tasks) | 56 | 97 | 93 |
| Claude Opus 5.5 (medium) | Anthropic | 85.4 | Not run: Vision agents (all 64 tasks) | 57 | 96 | 94 |
| Claude Sonnet 5.5 (high) | Anthropic | 84.3 | Not run: Vision agents (all 64 tasks) | 59 | 99 | 94 |
| Claude Haiku 5.5 (high) | Anthropic | 81.0 | Not run: Vision agents (all 64 tasks) | 47 | 91 | 92 |
| Claude Haiku 5.5 (medium) | Anthropic | 79.2 | Not run: Vision agents (all 64 tasks) | 45 | 94 | 92 |
| Claude Sonnet 5.5 (medium) | Anthropic | 78.9 | Not run: Vision agents (all 64 tasks) | 56 | 96 | 93 |
| Kimi K3 (Reasoning) Open weights | Moonshot AI | 78.6 | Not run: Vision agents (all 64 tasks) | 48 | 92 | 92 |
| GLM-5.3-Flash (low) Open weights | Z.ai | 71.1 | Not run: Vision agents (all 64 tasks) | Not run: 14 of 136 questions | 94 | 85 |
| GLM-5.3-Flash (max) Open weights | Z.ai | 71.0 | Not run: Vision agents (all 64 tasks) | Incomplete: 14 of 136 questions rejected by the provider | 84 | 92 |
| Gemma 4 26B A4B (Reasoning) Open weights | 64.4 | Not run: Vision agents (all 64 tasks) | 43 | 75 | 89 | |
| Mistral Large 3 (Non-reasoning) Open weights | Mistral AI | 44.5 | Not run: Vision agents (all 64 tasks) | Not run | 47 | 89 |
| GPT-6 Sol (medium) | OpenAI | Not yet ranked | Not run: Earth-Bench V2 (44 of 117 tasks); GeoBenchX (134 of 150 tasks); GS-QA2 (53 of 221 tasks); Vision agents (42 of 64 tasks) | Not run | Not run: GeoAgentBench (21 of 50 tasks); GeoAnalystBench (all 19 tasks) | Not run |
| Claude Sonnet 4.6 (high) | Anthropic | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 98 | Not run: 845 of 1,685 questions |
| GPT-5.6 Sol (medium) | OpenAI | Not yet ranked | Not run: Earth-Bench V2 (all 117 tasks); GeoBenchX (all 150 tasks); GS-QA2 (202 of 221 tasks); Vision agents (all 64 tasks) | Not run | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | Not run |
| Gemini 3.1 Flash-Lite (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GeoBenchX (2 of 150 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | Not run: GeoAgentBench (2 of 50 tasks) | Not run | |
| Gemini 3.1 Flash-Lite (low) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 58 | Not run | |
| Gemini 3.1 Flash-Lite (minimal) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 42 | Not run | |
| Gemini 3.1 Pro (preview) (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 90 | Not run | |
| Gemini 3.1 Pro (preview) (low) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 81 | Not run | |
| Gemini 3.8 Flash (high) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 53 | Not run | |
| Gemini 3.8 Flash (low) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run | 94 | Not run | |
| Kimi K2.6 (Reasoning) Open weights | Moonshot AI | Not yet ranked | Not run: Vision agents (all 64 tasks) | 44 | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 92 |
| Kimi K2.7 Code (Reasoning) Open weights | Moonshot AI | Not yet ranked | Not run: Vision agents (all 64 tasks) | 44 | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 92 |
| Mistral Small 3.1 24B (Non-reasoning) Open weights | Mistral AI | Not yet ranked | No agent entry | 33 | No agent entry | 82 |
| Qwen 3.8 27B (xhigh) Open weights | Alibaba (Qwen) | Not yet ranked | Not run: Earth-Bench V2 (90 of 117 tasks); GeoBenchX (all 150 tasks); GS-QA2 (all 221 tasks); Vision agents (all 64 tasks) | Not run: 8 of 136 questions | Not run: GeoAgentBench (2 of 50 tasks); GeoAnalystBench (14 of 19 tasks) | 91 |
| gpt-oss-120b (high) Open weights (No image input) | OpenAI | Not yet ranked | 43 | No image input | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 90 |
| gpt-oss-20b (high) Open weights (No image input) | OpenAI | Not yet ranked | 9 | No image input | Not run: GeoAgentBench (all 50 tasks); GeoAnalystBench (all 19 tasks) | 85 |