Axis Spatial

Search modelsMenu

Geospatial Agent Benchmarks · results of 30 October 2026

Which AI model is best at GIS and Earth observation work?

Claude Opus 5.5 leads the Geospatial Intelligence Index at 89 of 100, ahead of GPT-6.1 Sol at 86.

The indexes

Geospatial Intelligence Index

Text only view, 0 to 100 · Higher is better

Geospatial Intelligence IndexScores 0 to 100: Claude Opus 5.5 (xhigh): 89; GPT-6.1 Sol (xhigh): 86; Claude Sonnet 5.5 (xhigh): 84; Claude Haiku 5.5 (xhigh): 82; GPT-5.6 Luna (xhigh): 80; DeepSeek V4 Pro (max): 79; Gemini 3.8 Flash (low): 79; Kimi K3 (Reasoning): 79; GLM-5.2 (max): 78; GPT-5.6 Terra (high): 78; DeepSeek V4 Flash (max): 77; Qwen 3.8 27B (xhigh): 75; Nemotron 3 Ultra (high): 74; GLM-5.3-Flash (max): 71; Gemma 4 26B A4B (Reasoning): 64; Gemini 3.1 Flash-Lite (high): 63; MiniMax M3.1 Flash (max): 60; Nemotron 3.5 Super VL 120B (Reasoning): 59; Step 5 Preview (Reasoning): 56; gpt-oss-120b (high): 52; Mistral Large 3 (Non-reasoning): 44; Nemotron 3 Nano Omni 30B (Reasoning): 32; Nemotron 3.5 Lightning 30B (Reasoning): 27; gpt-oss-20b (high): 22; Llama 3.3 70B (Non-reasoning): 200255075100Claude Opus 5.5 (xhigh)89GPT-6.1 Sol (xhigh)86Claude Sonnet 5.5 (xhigh)84Claude Haiku 5.5 (xhigh)82GPT-5.6 Luna (xhigh)80DeepSeek V4 Pro (max)79Gemini 3.8 Flash (low)79Kimi K3 (Reasoning)79GLM-5.2 (max)78GPT-5.6 Terra (high)78DeepSeek V4 Flash (max)77Qwen 3.8 27B (xhigh)75Nemotron 3 Ultra (high)74GLM-5.3-Flash (max)71Gemma 4 26B A4B (Reasoning)64Gemini 3.1 Flash-Lite (high)63MiniMax M3.1 Flash (max)60Nemotron 3.5 Super VL 120B (Reasoning)59Step 5 Preview (Reasoning)56gpt-oss-120b (high)52Mistral Large 3 (Non-reasoning)44Nemotron 3 Nano Omni 30B (Reasoning)32Nemotron 3.5 Lightning 30B (Reasoning)27gpt-oss-20b (high)22Llama 3.3 70B (Non-reasoning)20
Data table: Geospatial Intelligence Index
Geospatial Intelligence Index
ModelGeospatial Intelligence Index
Claude Opus 5.5 (xhigh)89
GPT-6.1 Sol (xhigh)86
GPT-6 Sol (high)85
Claude Opus 5 (high)84
Claude Sonnet 5.5 (xhigh)84
Claude Opus 4.8 (high)83
Claude Haiku 5.5 (xhigh)82
GPT-5.6 Sol (high)81
Claude Sonnet 5 (high)81
Claude Opus 4.7 (high)81
GPT-5.6 Luna (xhigh)80
GPT-6 Luna (xhigh)80
DeepSeek V4 Pro (max) Open weights79
Gemini 3.8 Flash (low)79
Kimi K3 (Reasoning) Open weights79
GLM-5.2 (max) Open weights78
GPT-5.6 Terra (high)78
DeepSeek V4 Flash (max) Open weights77
GLM-5.3 (max) Open weights77
Qwen 3.8 27B (xhigh) Open weights75
Nemotron 3 Ultra (high) Open weights74
GLM-5.3-Flash (max) Open weights71
Claude Haiku 4.5 (Reasoning)66
Gemma 4 26B A4B (Reasoning) Open weights64
Gemini 3.1 Flash-Lite (high)63
MiniMax M3.1 Flash (max) Open weights60
Nemotron 3.5 Super VL 120B (Reasoning) Open weights59
Step 5 Preview (Reasoning) Open weights56
gpt-oss-120b (high) Open weights52
GLM-4.7-Flash (Reasoning) Open weights50
Mistral Large 3 (Non-reasoning) Open weights44
Nemotron 3 Nano Omni 30B (Reasoning) Open weights32
Nemotron 3.5 Lightning 30B (Reasoning) Open weights27
gpt-oss-20b (high) Open weights22
Llama 3.3 70B (Non-reasoning) Open weights20

Full ranking: Geospatial Intelligence Index

Geospatial Agent Index

Equal average of Earth-Bench V2, GeoBenchX and GS-QA2, 0 to 100 · Higher is better

Geospatial Agent IndexScores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 85; GPT-6.1 Sol (xhigh), Codex: 82; Claude Sonnet 5.5 (xhigh), Claude Code: 79; Claude Haiku 5.5 (xhigh), Claude Code: 76; GLM-5.3 (max), Claude Code: 73; GPT-5.6 Luna (xhigh), Codex: 73; Kimi K3 (Reasoning), Terminus-2: 72; Kimi K2.7 Code (Reasoning), Kimi Code: 72; GLM-5.3-Flash (max), Claude Code: 72; Gemini 3.8 Flash (low), Terminus-2: 71; DeepSeek V4 Pro (max), Terminus-2: 71; DeepSeek V4.1 Flash (max), Codex: 71; GPT-5.6 Terra (high), Codex: 69; Qwen 3.8 27B (xhigh), Qwen Code: 68; Nemotron 3 Ultra (high), Terminus-2: 68; Gemini 3.1 Flash-Lite (high), Terminus-2: 56; Gemma 4 26B A4B (Reasoning), Terminus-2: 56; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 52; MiniMax M3.1 Flash (max), Terminus-2: 52; gpt-oss-120b (high), Terminus-2: 46; Step 5 Preview (Reasoning), Terminus-2: 40; gpt-oss-20b (high), Terminus-2: 33; Mistral Large 3 (Non-reasoning), Terminus-2: 33; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 21; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 13; Llama 3.3 70B (Non-reasoning), Terminus-2: 80255075100Claude Opus 5.5 (xhigh)Claude Code85GPT-6.1 Sol (xhigh)Codex82Claude Sonnet 5.5 (xhigh)Claude Code79Claude Haiku 5.5 (xhigh)Claude Code76GLM-5.3 (max)Claude Code73GPT-5.6 Luna (xhigh)Codex73Kimi K3 (Reasoning)Terminus-272Kimi K2.7 Code (Reasoning)Kimi Code72GLM-5.3-Flash (max)Claude Code72Gemini 3.8 Flash (low)Terminus-271DeepSeek V4 Pro (max)Terminus-271DeepSeek V4.1 Flash (max)Codex71GPT-5.6 Terra (high)Codex69Qwen 3.8 27B (xhigh)Qwen Code68Nemotron 3 Ultra (high)Terminus-268Gemini 3.1 Flash-Lite (high)Terminus-256Gemma 4 26B A4B (Reasoning)Terminus-256Nemotron 3.5 Super VL 120B (Reasoning)Terminus-252MiniMax M3.1 Flash (max)Terminus-252gpt-oss-120b (high)Terminus-246Step 5 Preview (Reasoning)Terminus-240gpt-oss-20b (high)Terminus-233Mistral Large 3 (Non-reasoning)Terminus-233Nemotron 3 Nano Omni 30B (Reasoning)Terminus-221Nemotron 3.5 Lightning 30B (Reasoning)Terminus-213Llama 3.3 70B (Non-reasoning)Terminus-28
Data table: Geospatial Agent Index
Geospatial Agent Index
ModelDeveloperGeospatial Agent Index
Claude Opus 5.5 (xhigh), Claude CodeAnthropic85
Claude Opus 5.5 (high), Claude CodeAnthropic82
GPT-6.1 Sol (xhigh), CodexOpenAI82
GPT-6.1 Sol (medium), CodexOpenAI82
Claude Opus 5.5 (medium), Claude CodeAnthropic80
GPT-6.1 Sol (high), CodexOpenAI80
Claude Opus 5 (high), Claude CodeAnthropic80
GPT-6 Sol (high), CodexOpenAI80
Claude Sonnet 5.5 (xhigh), Claude CodeAnthropic79
Claude Sonnet 5.5 (high), Claude CodeAnthropic78
Claude Opus 4.8 (high), Claude CodeAnthropic76
Claude Haiku 5.5 (xhigh), Claude CodeAnthropic76
Claude Haiku 5.5 (high), Claude CodeAnthropic76
GPT-5.6 Sol (high), CodexOpenAI75
Claude Sonnet 5 (high), Claude CodeAnthropic74
Claude Opus 5.5 (low), Claude CodeAnthropic74
Claude Opus 4.7 (high), Claude CodeAnthropic74
GLM-5.3 (max), Claude CodeZ.ai73
GLM-5.2 (max), Claude CodeZ.ai73
GPT-5.6 Luna (xhigh), CodexOpenAI73
GPT-6 Luna (xhigh), CodexOpenAI73
Claude Haiku 5.5 (medium), Claude CodeAnthropic72
Kimi K3 (Reasoning), Terminus-2Moonshot AI72
DeepSeek V4 Flash (high), Terminus-2DeepSeek72
Kimi K2.7 Code (Reasoning), Kimi CodeMoonshot AI72
GLM-5.2 (max), Terminus-2Z.ai72
GPT-5.6 Luna (high), CodexOpenAI72
GLM-5.3-Flash (max), Claude CodeZ.ai72
GPT-6 Luna (high), CodexOpenAI71
Gemini 3.8 Flash (low), Terminus-2Google71
Kimi K2.7 Code (Reasoning), Terminus-2Moonshot AI71
DeepSeek V4 Pro (max), Terminus-2DeepSeek71
Claude Sonnet 5.5 (medium), Claude CodeAnthropic71
GLM-5.2 (high), Terminus-2Z.ai71
DeepSeek V4.1 Flash (max), CodexDeepSeek71
GLM-5.3 (max), Terminus-2Z.ai70
DeepSeek V4 Pro (high), Terminus-2DeepSeek70
Claude Haiku 5.5 (low), Claude CodeAnthropic70
Kimi K2.6 (Reasoning), Terminus-2Moonshot AI70
GPT-5.6 Terra (high), CodexOpenAI69
Claude Sonnet 5.5 (low), Claude CodeAnthropic69
DeepSeek V4 Flash (max), Terminus-2DeepSeek69
Kimi K2.6 (Reasoning), Kimi CodeMoonshot AI69
Qwen 3.8 27B (xhigh), Qwen CodeAlibaba (Qwen)68
Nemotron 3 Ultra (high), Terminus-2NVIDIA68
GPT-5.6 Luna (medium), CodexOpenAI66
Qwen 3.8 27B (xhigh), Terminus-2Alibaba (Qwen)66
GLM-5.3 (low), Terminus-2Z.ai66
GLM-5.2 (Non-reasoning), Terminus-2Z.ai63
GLM-5.3-Flash (max), Terminus-2Z.ai63
GLM-5.3-Flash (low), Terminus-2Z.ai62
GPT-5.6 Luna (low), CodexOpenAI61
DeepSeek V4 Flash (high), CodexDeepSeek59
Claude Haiku 4.5 (Reasoning), Claude CodeAnthropic58
GPT-6 Luna (medium), CodexOpenAI57
Gemini 3.1 Flash-Lite (high), Terminus-2Google56
Gemma 4 26B A4B (Reasoning), Terminus-2Google56
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA52
MiniMax M3.1 Flash (max), Terminus-2MiniMax52
GPT-6 Luna (low), CodexOpenAI51
Gemini 3.1 Flash-Lite (low), Terminus-2Google48
gpt-oss-120b (high), Terminus-2OpenAI46
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai43
gpt-oss-120b (high), CodexOpenAI43
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai41
Step 5 Preview (Reasoning), Terminus-2StepFun40
gpt-oss-20b (high), Terminus-2OpenAI33
Mistral Large 3 (Non-reasoning), Terminus-2Mistral AI33
Gemini 3.1 Flash-Lite (minimal), Terminus-2Google32
gpt-oss-120b (low), Terminus-2OpenAI31
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIA21
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIA13
gpt-oss-20b (high), CodexOpenAI9
Llama 3.3 70B (Non-reasoning), Terminus-2Meta8

Full ranking: Geospatial Agent Index

Vision

Questions answered from satellite and aerial images, with no tools, 0 to 100 · Higher is better

VisionScores 0 to 100: Claude Opus 5.5 (xhigh): 59; GPT-5.6 Sol (high): 58; Claude Sonnet 5.5 (xhigh): 57; GPT-6.1 Sol (xhigh): 56; GPT-5.6 Terra (high): 54; Qwen 3.8 27B (xhigh): 53; Claude Haiku 5.5 (xhigh): 52; GPT-5.6 Luna (xhigh): 52; Gemini 3.8 Flash (low): 50; Kimi K3 (Reasoning): 48; GLM-5.3-Flash (max): 44; Kimi K2.7 Code (Reasoning): 44; Gemma 4 26B A4B (Reasoning): 43; Gemini 3.1 Flash-Lite (high): 38; Mistral Large 3 (Non-reasoning): 33; Mistral Small 3.1 24B (Non-reasoning): 330255075100Claude Opus 5.5 (xhigh)59GPT-5.6 Sol (high)58Claude Sonnet 5.5 (xhigh)57GPT-6.1 Sol (xhigh)56GPT-5.6 Terra (high)54Qwen 3.8 27B (xhigh)53Claude Haiku 5.5 (xhigh)52GPT-5.6 Luna (xhigh)52Gemini 3.8 Flash (low)50Kimi K3 (Reasoning)48GLM-5.3-Flash (max)44Kimi K2.7 Code (Reasoning)44Gemma 4 26B A4B (Reasoning)43Gemini 3.1 Flash-Lite (high)38Mistral Large 3 (Non-reasoning)33Mistral Small 3.1 24B (Non-reasoning)33
Data table: Vision
Vision
ModelVision
Claude Opus 5.5 (xhigh)59
GPT-5.6 Sol (high)58
Claude Sonnet 5.5 (xhigh)57
GPT-6.1 Sol (xhigh)56
GPT-6 Sol (high)56
GPT-5.6 Terra (high)54
Claude Opus 5 (high)53
Qwen 3.8 27B (xhigh) Open weights53
Claude Haiku 5.5 (xhigh)52
GPT-5.6 Luna (xhigh)52
GPT-6 Luna (xhigh)50
Claude Opus 4.8 (high)50
Gemini 3.8 Flash (low)50
Kimi K3 (Reasoning) Open weights48
Claude Opus 4.7 (high)47
Claude Sonnet 5 (high)47
Kimi K2.6 (Reasoning) Open weights44
GLM-5.3-Flash (max) Open weights44
Kimi K2.7 Code (Reasoning) Open weights44
Gemma 4 26B A4B (Reasoning) Open weights43
Gemini 3.1 Flash-Lite (high)38
Claude Haiku 4.5 (Reasoning)34
Mistral Large 3 (Non-reasoning) Open weights33
Mistral Small 3.1 24B (Non-reasoning) Open weights33

Full ranking: Vision

Geospatial knowledge

Share of questions answered correctly, with no tools, 0 to 100 · Higher is better

Geospatial knowledgeScores 0 to 100: Claude Opus 5.5 (xhigh): 94; Claude Sonnet 5.5 (xhigh): 94; Gemini 3.8 Flash (low): 94; GPT-5.6 Sol (high): 93; GPT-5.6 Terra (high): 92; Kimi K2.7 Code (Reasoning): 92; Kimi K3 (Reasoning): 92; DeepSeek V4 Pro (max): 92; Nemotron 3.5 Super VL 120B (Reasoning): 92; GLM-5.3 (max): 92; Claude Haiku 5.5 (xhigh): 92; GLM-5.3-Flash (max): 92; Step 5 Preview (Reasoning): 92; DeepSeek V4 Flash (max): 91; GPT-6 Luna (xhigh): 91; Qwen 3.8 27B (xhigh): 91; Nemotron 3 Ultra (high): 91; Gemini 3.1 Flash-Lite (high): 90; gpt-oss-120b (high): 90; Mistral Large 3 (Non-reasoning): 89; Gemma 4 26B A4B (Reasoning): 89; gpt-oss-20b (high): 85; Llama 3.3 70B (Non-reasoning): 82; Mistral Small 3.1 24B (Non-reasoning): 820255075100Claude Opus 5.5 (xhigh)94Claude Sonnet 5.5 (xhigh)94Gemini 3.8 Flash (low)94GPT-5.6 Sol (high)93GPT-5.6 Terra (high)92Kimi K2.7 Code (Reasoning)92Kimi K3 (Reasoning)92DeepSeek V4 Pro (max)92Nemotron 3.5 Super VL 120B (Reasoning)92GLM-5.3 (max)92Claude Haiku 5.5 (xhigh)92GLM-5.3-Flash (max)92Step 5 Preview (Reasoning)92DeepSeek V4 Flash (max)91GPT-6 Luna (xhigh)91Qwen 3.8 27B (xhigh)91Nemotron 3 Ultra (high)91Gemini 3.1 Flash-Lite (high)90gpt-oss-120b (high)90Mistral Large 3 (Non-reasoning)89Gemma 4 26B A4B (Reasoning)89gpt-oss-20b (high)85Llama 3.3 70B (Non-reasoning)82Mistral Small 3.1 24B (Non-reasoning)82
Data table: Geospatial knowledge
Geospatial knowledge
ModelGeospatial knowledge
Claude Opus 5.5 (xhigh)94
Claude Sonnet 5.5 (xhigh)94
Gemini 3.8 Flash (low)94
Claude Opus 4.8 (high)93
Claude Opus 5 (high)93
Claude Opus 4.7 (high)93
GPT-5.6 Sol (high)93
GPT-6.1 Sol (xhigh)93
GPT-5.6 Terra (high)92
Kimi K2.7 Code (Reasoning) Open weights92
Kimi K3 (Reasoning) Open weights92
DeepSeek V4 Pro (max) Open weights92
Nemotron 3.5 Super VL 120B (Reasoning) Open weights92
Claude Haiku 5.5 (xhigh)92
GLM-5.3 (max) Open weights92
Kimi K2.6 (Reasoning) Open weights92
GLM-5.3-Flash (max) Open weights92
Step 5 Preview (Reasoning) Open weights92
DeepSeek V4 Flash (max) Open weights91
Claude Sonnet 5 (high)91
GPT-6 Luna (xhigh)91
Qwen 3.8 27B (xhigh) Open weights91
GPT-6 Sol (high)91
Nemotron 3 Ultra (high) Open weights91
GPT-5.6 Luna (xhigh)91
Gemini 3.1 Flash-Lite (high)90
gpt-oss-120b (high) Open weights90
Claude Haiku 4.5 (Reasoning)89
GLM-5.2 (max) Open weights89
Mistral Large 3 (Non-reasoning) Open weights89
Gemma 4 26B A4B (Reasoning) Open weights89
GLM-4.7-Flash (Reasoning) Open weights86
gpt-oss-20b (high) Open weights85
Llama 3.3 70B (Non-reasoning) Open weights82
Mistral Small 3.1 24B (Non-reasoning) Open weights82

Full ranking: Geospatial knowledge

Find a model or compare two

Every model · Every comparison

Picks

Geospatial Agent Index, 0 to 100. Picks for each kind of job

Popular comparisons

All comparisons

What’s new

All updates · Atom feed

The tasks behind the scores

World map of the places the tasks are setWhere the tasks areThe places on Earth the tasks are set; the larger a circle, the more tasks there

What they read

How we benchmark

  • Every model gets the same real GIS and Earth observation tasks, data and limits.
  • Each task is scored the same way for every model, on one attempt.
  • Costs are at public API list prices, and every change to the method is in its version history.

The methodology

Frequently asked questions

What do these benchmarks measure?
How well AI models do geospatial (GIS and Earth observation) work: agent tasks, reading satellite and aerial images, guided workflows and geospatial knowledge. Every model gets the same tasks and limits and is scored the same way.
Which index should I look at?
The Geospatial Intelligence Index ranks each model once, across all four groups. The Geospatial Agent Index compares entries: one model in one harness at one reasoning level, on agent tasks. Vision and Knowledge show those groups on their own.
How do I hear about new results?
Updates lists every new model, release and correction, with an Atom feed, or sign up by email above.