Axis Spatial

Search modelsMenu

Z.ai: geospatial agent results

Helps you choose the best Z.ai model for geospatial work: GIS and Earth observation agent tasks, vision and knowledge, with cost and time.

Overview

Most intelligent

  1. #1GLM-5.2 (high) Open weights79
  2. #2GLM-5.2 (max) Open weights78
  3. #3GLM-5.3 (max) Open weights77
  4. #4GLM-5.3 (low) Open weights75
  5. #5GLM-5.3-Flash (low) Open weights71

Geospatial Intelligence Index, Text only

Lowest cost

  1. #1GLM-4.7-Flash (Reasoning), Terminus-2$0.037
  2. #2GLM-5.3-Flash (low), Terminus-2$0.048
  3. #3GLM-5.3-Flash (max), Claude Code$0.093
  4. #4GLM-5.3-Flash (max), Terminus-2$0.094
  5. #5GLM-5.2 (Non-reasoning), Terminus-2$0.099

Cost per agent task, at list prices

Fastest

  1. #1GLM-5.2 (Non-reasoning), Terminus-21.5 min
  2. #2GLM-5.3 (low), Terminus-22.7 min
  3. #3GLM-4.7-Flash (Reasoning), Claude Code3.8 min
  4. #4GLM-5.3 (max), Terminus-24.2 min
  5. #5GLM-5.3-Flash (low), Terminus-24.2 min

Time per agent task, at 100 output tokens/s

  • For intelligence, the top Z.ai models are GLM-5.2 (high) (79), GLM-5.2 (max) (78) and GLM-5.3 (max) (77) on the Geospatial Intelligence Index (Text only).
  • For cost, the cheapest per agent task are GLM-4.7-Flash (Reasoning) ($0.037), GLM-5.3-Flash (low) ($0.048) and GLM-5.3-Flash (max) ($0.093).
  • For speed, the fastest per agent task are GLM-5.2 (Non-reasoning) (1.5 min), GLM-5.3 (low) (2.7 min) and GLM-4.7-Flash (Reasoning) (3.8 min).
  • For context window, the largest are GLM-5.3-Flash: 1M tokens, GLM-5.3: 1M tokens and GLM-5.2: 262K tokens.

Intelligence

Geospatial Intelligence Index, Text only · Higher is better

IntelligenceScores 0 to 100: GLM-5.2 (high): 79; GLM-5.2 (max): 78; GLM-5.3 (max): 77; GLM-5.3 (low): 75; GLM-5.3-Flash (low): 71; GLM-5.3-Flash (max): 71; GLM-5.2 (Non-reasoning): 71; GLM-4.7-Flash (Reasoning): 500255075100GLM-5.2 (high)79GLM-5.2 (max)78GLM-5.3 (max)77GLM-5.3 (low)75GLM-5.3-Flash (low)71GLM-5.3-Flash (max)71GLM-5.2 (Non-reasoning)71GLM-4.7-Flash (Reasoning)50
Data table: Intelligence
Intelligence
ModelDeveloperIntelligence
GLM-5.2 (high)Z.ai79
GLM-5.2 (max)Z.ai78
GLM-5.3 (max)Z.ai77
GLM-5.3 (low)Z.ai75
GLM-5.3-Flash (low)Z.ai71
GLM-5.3-Flash (max)Z.ai71
GLM-5.2 (Non-reasoning)Z.ai71
GLM-4.7-Flash (Reasoning)Z.ai50

Cost per task

USD per agent task · Lower is better

Cost per taskCost per task in US dollars: GLM-4.7-Flash (Reasoning), Terminus-2: $0.037; GLM-5.3-Flash (low), Terminus-2: $0.048; GLM-5.3-Flash (max), Claude Code: $0.093; GLM-5.3-Flash (max), Terminus-2: $0.094; GLM-5.2 (Non-reasoning), Terminus-2: $0.099; GLM-5.2 (high), Terminus-2: $0.22; GLM-4.7-Flash (Reasoning), Claude Code: $0.27; GLM-5.3 (low), Terminus-2: $0.27; GLM-5.2 (max), Terminus-2: $0.32; GLM-5.2 (max), Claude Code: $0.34; GLM-5.3 (max), Terminus-2: $0.38; GLM-5.3 (max), Claude Code: $0.48$0$0.15$0.30$0.45$0.60GLM-4.7-Flash (Reasoning)Terminus-2$0.037GLM-5.3-Flash (low)Terminus-2$0.048GLM-5.3-Flash (max)Claude Code$0.093GLM-5.3-Flash (max)Terminus-2$0.094GLM-5.2 (Non-reasoning)Terminus-2$0.099GLM-5.2 (high)Terminus-2$0.22GLM-4.7-Flash (Reasoning)Claude Code$0.27GLM-5.3 (low)Terminus-2$0.27GLM-5.2 (max)Terminus-2$0.32GLM-5.2 (max)Claude Code$0.34GLM-5.3 (max)Terminus-2$0.38GLM-5.3 (max)Claude Code$0.48
Data table: Cost per task
Cost per task
ModelDeveloperCost per task
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai$0.037
GLM-5.3-Flash (low), Terminus-2Z.ai$0.048
GLM-5.3-Flash (max), Claude CodeZ.ai$0.093
GLM-5.3-Flash (max), Terminus-2Z.ai$0.094
GLM-5.2 (Non-reasoning), Terminus-2Z.ai$0.099
GLM-5.2 (high), Terminus-2Z.ai$0.22
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai$0.27
GLM-5.3 (low), Terminus-2Z.ai$0.27
GLM-5.2 (max), Terminus-2Z.ai$0.32
GLM-5.2 (max), Claude CodeZ.ai$0.34
GLM-5.3 (max), Terminus-2Z.ai$0.38
GLM-5.3 (max), Claude CodeZ.ai$0.48

Time per task

Minutes per agent task · Lower is better

Time per taskTime per task: GLM-5.2 (Non-reasoning), Terminus-2: 1.5 min; GLM-5.3 (low), Terminus-2: 2.7 min; GLM-4.7-Flash (Reasoning), Claude Code: 3.8 min; GLM-5.3 (max), Terminus-2: 4.2 min; GLM-5.3-Flash (low), Terminus-2: 4.2 min; GLM-5.3 (max), Claude Code: 4.3 min; GLM-5.2 (high), Terminus-2: 4.7 min; GLM-4.7-Flash (Reasoning), Terminus-2: 5.4 min; GLM-5.2 (max), Terminus-2: 6.1 min; GLM-5.2 (max), Claude Code: 6.3 min; GLM-5.3-Flash (max), Claude Code: 6.5 min; GLM-5.3-Flash (max), Terminus-2: 8.5 min02.557.510GLM-5.2 (Non-reasoning)Terminus-21.5 minGLM-5.3 (low)Terminus-22.7 minGLM-4.7-Flash (Reasoning)Claude Code3.8 minGLM-5.3 (max)Terminus-24.2 minGLM-5.3-Flash (low)Terminus-24.2 minGLM-5.3 (max)Claude Code4.3 minGLM-5.2 (high)Terminus-24.7 minGLM-4.7-Flash (Reasoning)Terminus-25.4 minGLM-5.2 (max)Terminus-26.1 minGLM-5.2 (max)Claude Code6.3 minGLM-5.3-Flash (max)Claude Code6.5 minGLM-5.3-Flash (max)Terminus-28.5 min
Data table: Time per task
Time per task
ModelDeveloperTime per task (at 100 output tokens/s)
GLM-5.2 (Non-reasoning), Terminus-2Z.ai1.5 min
GLM-5.3 (low), Terminus-2Z.ai2.7 min
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai3.8 min
GLM-5.3 (max), Terminus-2Z.ai4.2 min
GLM-5.3-Flash (low), Terminus-2Z.ai4.2 min
GLM-5.3 (max), Claude CodeZ.ai4.3 min
GLM-5.2 (high), Terminus-2Z.ai4.7 min
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai5.4 min
GLM-5.2 (max), Terminus-2Z.ai6.1 min
GLM-5.2 (max), Claude CodeZ.ai6.3 min
GLM-5.3-Flash (max), Claude CodeZ.ai6.5 min
GLM-5.3-Flash (max), Terminus-2Z.ai8.5 min

Models

GLM-5.3

Open weightsText only

1M token context · Released 18 August 2026

$1.40 input · $4.40 output per 1M tokens

GLM-5.2

Open weightsText only

262K token context · Released 16 June 2026

$1.40 input · $4.40 output per 1M tokens

GLM-5.3-Flash

Open weightsReads images

1M token context · Released 26 August 2026

$0.15 input · $0.50 output per 1M tokens

GLM-4.7-Flash

Open weightsText only

131K token context · Released 19 January 2026

$0.060 input · $0.40 output per 1M tokens

Intelligence

Text only

Geospatial Intelligence Index: Text only

Weighted average of agent tasks, guided workflows and geospatial knowledge, 0 to 100 · Higher is better

Geospatial Intelligence Index: Text onlyScores 0 to 100: Claude Opus 5.5 (xhigh): 89; GPT-6.1 Sol (medium): 86; DeepSeek V4 Pro (max): 79; GLM-5.2 (high): 79; GLM-5.2 (max): 78; GLM-5.3 (max): 77; GLM-5.3 (low): 75; GLM-5.3-Flash (low): 71; GLM-5.3-Flash (max): 71; GLM-5.2 (Non-reasoning): 71; GLM-4.7-Flash (Reasoning): 500255075100Claude Opus 5.5 (xhigh)89GPT-6.1 Sol (medium)86DeepSeek V4 Pro (max)79GLM-5.2 (high)79GLM-5.2 (max)78GLM-5.3 (max)77GLM-5.3 (low)75GLM-5.3-Flash (low)71GLM-5.3-Flash (max)71GLM-5.2 (Non-reasoning)71GLM-4.7-Flash (Reasoning)50
How to read this chart

What this metric means. A weighted average of group scores, each 0 to 100: Agent tasks 4/6, Guided workflows 1/6, Geospatial knowledge 1/6.

Bars. Every Z.ai model with a result, highlighted, and the strongest models from other labs, faded, for context. Provisional: not yet ranked.

Data table: Geospatial Intelligence Index: Text only
Geospatial Intelligence Index: Text only
ModelDeveloperGeospatial Intelligence Index: Text only
Claude Opus 5.5 (xhigh)Anthropic89
GPT-6.1 Sol (medium)OpenAI86
DeepSeek V4 Pro (max)DeepSeek79
GLM-5.2 (high)Z.ai79
GLM-5.2 (max)Z.ai78
GLM-5.3 (max)Z.ai77
GLM-5.3 (low)Z.ai75
GLM-5.3-Flash (low)Z.ai71
GLM-5.3-Flash (max)Z.ai71
GLM-5.2 (Non-reasoning)Z.ai71
GLM-4.7-Flash (Reasoning)Z.ai50

Text + vision

No Z.ai model has a Text + vision score yet: that view also needs Vision agents and Vision results.

Agent tasks

Text only

Geospatial Agent Index: Text only

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

Geospatial Agent Index: Text onlyScores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 85; GPT-6.1 Sol (xhigh), Codex: 82; GLM-5.3 (max), Claude Code: 73; GLM-5.2 (max), Claude Code: 73; Kimi K3 (Reasoning), Terminus-2: 72; GLM-5.2 (max), Terminus-2: 72; GLM-5.3-Flash (max), Claude Code: 72; GLM-5.2 (high), Terminus-2: 71; GLM-5.3 (max), Terminus-2: 70; GLM-5.3 (low), Terminus-2: 66; GLM-5.2 (Non-reasoning), Terminus-2: 63; GLM-5.3-Flash (max), Terminus-2: 63; GLM-5.3-Flash (low), Terminus-2: 62; GLM-4.7-Flash (Reasoning), Terminus-2: 43; GLM-4.7-Flash (Reasoning), Claude Code: 410255075100Claude Opus 5.5 (xhigh)Claude Code85GPT-6.1 Sol (xhigh)Codex82GLM-5.3 (max)Claude Code73GLM-5.2 (max)Claude Code73Kimi K3 (Reasoning)Terminus-272GLM-5.2 (max)Terminus-272GLM-5.3-Flash (max)Claude Code72GLM-5.2 (high)Terminus-271GLM-5.3 (max)Terminus-270GLM-5.3 (low)Terminus-266GLM-5.2 (Non-reasoning)Terminus-263GLM-5.3-Flash (max)Terminus-263GLM-5.3-Flash (low)Terminus-262GLM-4.7-Flash (Reasoning)Terminus-243GLM-4.7-Flash (Reasoning)Claude Code41
How to read this chart

Bars. Every Z.ai model with a result, highlighted, and the strongest models from other labs, faded, for context. Provisional: not yet ranked.

Data table: Geospatial Agent Index: Text only
Geospatial Agent Index: Text only
ModelDeveloperGeospatial Agent Index: Text only
Claude Opus 5.5 (xhigh), Claude CodeAnthropic85
GPT-6.1 Sol (xhigh), CodexOpenAI82
GLM-5.3 (max), Claude CodeZ.ai73
GLM-5.2 (max), Claude CodeZ.ai73
Kimi K3 (Reasoning), Terminus-2Moonshot AI72
GLM-5.2 (max), Terminus-2Z.ai72
GLM-5.3-Flash (max), Claude CodeZ.ai72
GLM-5.2 (high), Terminus-2Z.ai71
GLM-5.3 (max), Terminus-2Z.ai70
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
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai43
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai41

Text + vision

Geospatial Agent Index: Text + vision

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

Geospatial Agent Index: Text + visionScores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 88; GPT-6.1 Sol (xhigh), Codex: 84; GLM-5.3-Flash (max), Claude Code: 73; DeepSeek V4.1 Flash (max), Codex: 730255075100Claude Opus 5.5 (xhigh)Claude Code88GPT-6.1 Sol (xhigh)Codex84GLM-5.3-Flash (max)Claude Code73DeepSeek V4.1 Flash (max)Codex73
How to read this chart

Bars. Every Z.ai model with a result, highlighted, and the strongest models from other labs, faded, for context. Provisional: not yet ranked.

Data table: Geospatial Agent Index: Text + vision
Geospatial Agent Index: Text + vision
ModelDeveloperGeospatial Agent Index: Text + vision
Claude Opus 5.5 (xhigh), Claude CodeAnthropic88
GPT-6.1 Sol (xhigh), CodexOpenAI84
GLM-5.3-Flash (max), Claude CodeZ.ai73
DeepSeek V4.1 Flash (max), CodexDeepSeek73

Earth-Bench V2

Earth observation data analysis · 0 to 100 · Higher is better

Earth-Bench V2Scores 0 to 100: Claude Opus 5.5 (high), Claude Code: 79; GPT-6 Sol (high), Codex: 75; GLM-5.3 (max), Claude Code: 74; GLM-5.2 (max), Claude Code: 71; GLM-5.3-Flash (max), Claude Code: 69; GLM-5.2 (high), Terminus-2: 68; GLM-5.2 (max), Terminus-2: 68; GLM-5.3 (max), Terminus-2: 64; GLM-5.2 (Non-reasoning), Terminus-2: 58; GLM-5.3 (low), Terminus-2: 57; GLM-5.3-Flash (low), Terminus-2: 53; GLM-5.3-Flash (max), Terminus-2: 51; GLM-4.7-Flash (Reasoning), Terminus-2: 44; GLM-4.7-Flash (Reasoning), Claude Code: 380255075100Claude Opus 5.5 (high)Claude Code79GPT-6 Sol (high)Codex75GLM-5.3 (max)Claude Code74GLM-5.2 (max)Claude Code71GLM-5.3-Flash (max)Claude Code69GLM-5.2 (high)Terminus-268GLM-5.2 (max)Terminus-268GLM-5.3 (max)Terminus-264GLM-5.2 (Non-reasoning)Terminus-258GLM-5.3 (low)Terminus-257GLM-5.3-Flash (low)Terminus-253GLM-5.3-Flash (max)Terminus-251GLM-4.7-Flash (Reasoning)Terminus-244GLM-4.7-Flash (Reasoning)Claude Code38
Data table: Earth-Bench V2
Earth-Bench V2
ModelDeveloperEarth-Bench V2
Claude Opus 5.5 (high), Claude CodeAnthropic79
GPT-6 Sol (high), CodexOpenAI75
GLM-5.3 (max), Claude CodeZ.ai74
GLM-5.2 (max), Claude CodeZ.ai71
GLM-5.3-Flash (max), Claude CodeZ.ai69
GLM-5.2 (high), Terminus-2Z.ai68
GLM-5.2 (max), Terminus-2Z.ai68
GLM-5.3 (max), Terminus-2Z.ai64
GLM-5.2 (Non-reasoning), Terminus-2Z.ai58
GLM-5.3 (low), Terminus-2Z.ai57
GLM-5.3-Flash (low), Terminus-2Z.ai53
GLM-5.3-Flash (max), Terminus-2Z.ai51
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai44
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai38

GeoBenchX

Maps, numbers and lists from country and city data; some tasks should be declined · 0 to 100 · Higher is better

GeoBenchXScores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 83; GPT-6.1 Sol (medium), Codex: 79; GLM-5.3-Flash (max), Claude Code: 70; GLM-5.2 (max), Terminus-2: 67; GLM-5.2 (high), Terminus-2: 67; GLM-5.2 (max), Claude Code: 66; GLM-5.3 (max), Terminus-2: 65; GLM-5.3 (max), Claude Code: 65; GLM-5.2 (Non-reasoning), Terminus-2: 60; GLM-5.3 (low), Terminus-2: 60; GLM-5.3-Flash (low), Terminus-2: 57; GLM-5.3-Flash (max), Terminus-2: 57; GLM-4.7-Flash (Reasoning), Terminus-2: 42; GLM-4.7-Flash (Reasoning), Claude Code: 390255075100Claude Opus 5.5 (xhigh)Claude Code83GPT-6.1 Sol (medium)Codex79GLM-5.3-Flash (max)Claude Code70GLM-5.2 (max)Terminus-267GLM-5.2 (high)Terminus-267GLM-5.2 (max)Claude Code66GLM-5.3 (max)Terminus-265GLM-5.3 (max)Claude Code65GLM-5.2 (Non-reasoning)Terminus-260GLM-5.3 (low)Terminus-260GLM-5.3-Flash (low)Terminus-257GLM-5.3-Flash (max)Terminus-257GLM-4.7-Flash (Reasoning)Terminus-242GLM-4.7-Flash (Reasoning)Claude Code39
Data table: GeoBenchX
GeoBenchX
ModelDeveloperGeoBenchX
Claude Opus 5.5 (xhigh), Claude CodeAnthropic83
GPT-6.1 Sol (medium), CodexOpenAI79
GLM-5.3-Flash (max), Claude CodeZ.ai70
GLM-5.2 (max), Terminus-2Z.ai67
GLM-5.2 (high), Terminus-2Z.ai67
GLM-5.2 (max), Claude CodeZ.ai66
GLM-5.3 (max), Terminus-2Z.ai65
GLM-5.3 (max), Claude CodeZ.ai65
GLM-5.2 (Non-reasoning), Terminus-2Z.ai60
GLM-5.3 (low), Terminus-2Z.ai60
GLM-5.3-Flash (low), Terminus-2Z.ai57
GLM-5.3-Flash (max), Terminus-2Z.ai57
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai42
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai39

GS-QA2

Spatial data reasoning over map and elevation data · 0 to 100 · Higher is better

GS-QA2Scores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 94; GPT-6.1 Sol (xhigh), Codex: 94; GLM-5.2 (max), Claude Code: 83; GLM-5.3 (max), Terminus-2: 83; GLM-5.3 (max), Claude Code: 81; GLM-5.2 (max), Terminus-2: 80; GLM-5.3 (low), Terminus-2: 80; GLM-5.3-Flash (max), Terminus-2: 80; GLM-5.2 (high), Terminus-2: 78; GLM-5.3-Flash (max), Claude Code: 76; GLM-5.3-Flash (low), Terminus-2: 76; GLM-5.2 (Non-reasoning), Terminus-2: 72; GLM-4.7-Flash (Reasoning), Claude Code: 48; GLM-4.7-Flash (Reasoning), Terminus-2: 430255075100Claude Opus 5.5 (xhigh)Claude Code94GPT-6.1 Sol (xhigh)Codex94GLM-5.2 (max)Claude Code83GLM-5.3 (max)Terminus-283GLM-5.3 (max)Claude Code81GLM-5.2 (max)Terminus-280GLM-5.3 (low)Terminus-280GLM-5.3-Flash (max)Terminus-280GLM-5.2 (high)Terminus-278GLM-5.3-Flash (max)Claude Code76GLM-5.3-Flash (low)Terminus-276GLM-5.2 (Non-reasoning)Terminus-272GLM-4.7-Flash (Reasoning)Claude Code48GLM-4.7-Flash (Reasoning)Terminus-243
Data table: GS-QA2
GS-QA2
ModelDeveloperGS-QA2
Claude Opus 5.5 (xhigh), Claude CodeAnthropic94
GPT-6.1 Sol (xhigh), CodexOpenAI94
GLM-5.2 (max), Claude CodeZ.ai83
GLM-5.3 (max), Terminus-2Z.ai83
GLM-5.3 (max), Claude CodeZ.ai81
GLM-5.2 (max), Terminus-2Z.ai80
GLM-5.3 (low), Terminus-2Z.ai80
GLM-5.3-Flash (max), Terminus-2Z.ai80
GLM-5.2 (high), Terminus-2Z.ai78
GLM-5.3-Flash (max), Claude CodeZ.ai76
GLM-5.3-Flash (low), Terminus-2Z.ai76
GLM-5.2 (Non-reasoning), Terminus-2Z.ai72
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai48
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai43

GeoAgentBench

Guided workflows: multi-step spatial execution · 0 to 100 · Higher is better

GeoAgentBenchScores 0 to 100: Claude Opus 5.5 (high), Claude Code: 100; GPT-6.1 Sol (medium), Codex: 100; GLM-5.3-Flash (max), Claude Code: 98; GLM-5.3 (max), Claude Code: 98; GLM-5.2 (high), Terminus-2: 94; GLM-5.2 (max), Terminus-2: 94; GLM-5.3-Flash (low), Terminus-2: 94; GLM-5.3 (max), Terminus-2: 94; GLM-5.2 (Non-reasoning), Terminus-2: 92; GLM-5.3 (low), Terminus-2: 92; GLM-5.3-Flash (max), Terminus-2: 84; GLM-4.7-Flash (Reasoning), Terminus-2: 42; GLM-4.7-Flash (Reasoning), Claude Code: 300255075100Claude Opus 5.5 (high)Claude Code100GPT-6.1 Sol (medium)Codex100GLM-5.3-Flash (max)Claude Code98GLM-5.3 (max)Claude Code98GLM-5.2 (high)Terminus-294GLM-5.2 (max)Terminus-294GLM-5.3-Flash (low)Terminus-294GLM-5.3 (max)Terminus-294GLM-5.2 (Non-reasoning)Terminus-292GLM-5.3 (low)Terminus-292GLM-5.3-Flash (max)Terminus-284GLM-4.7-Flash (Reasoning)Terminus-242GLM-4.7-Flash (Reasoning)Claude Code30
Data table: GeoAgentBench
GeoAgentBench
ModelDeveloperGeoAgentBench
Claude Opus 5.5 (high), Claude CodeAnthropic100
GPT-6.1 Sol (medium), CodexOpenAI100
GLM-5.3-Flash (max), Claude CodeZ.ai98
GLM-5.3 (max), Claude CodeZ.ai98
GLM-5.2 (high), Terminus-2Z.ai94
GLM-5.2 (max), Terminus-2Z.ai94
GLM-5.3-Flash (low), Terminus-2Z.ai94
GLM-5.3 (max), Terminus-2Z.ai94
GLM-5.2 (Non-reasoning), Terminus-2Z.ai92
GLM-5.3 (low), Terminus-2Z.ai92
GLM-5.3-Flash (max), Terminus-2Z.ai84
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai42
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai30

GeoAnalystBench

Guided workflows: spatial workflow and code generation · 0 to 100 · Higher is better

GeoAnalystBenchScores 0 to 100: GLM-5.2 (high), Terminus-2: 100; GLM-5.3-Flash (max), Claude Code: 100; DeepSeek V4 Pro (max), Terminus-2: 100; Claude Opus 5.5 (high), Claude Code: 100; GLM-5.2 (max), Terminus-2: 95; GLM-5.3-Flash (low), Terminus-2: 95; GLM-5.3 (low), Terminus-2: 95; GLM-5.3 (max), Claude Code: 95; GLM-5.3-Flash (max), Terminus-2: 84; GLM-5.3 (max), Terminus-2: 84; GLM-5.2 (Non-reasoning), Terminus-2: 74; GLM-4.7-Flash (Reasoning), Terminus-2: 37; GLM-4.7-Flash (Reasoning), Claude Code: 260255075100GLM-5.2 (high)Terminus-2100GLM-5.3-Flash (max)Claude Code100DeepSeek V4 Pro (max)Terminus-2100Claude Opus 5.5 (high)Claude Code100GLM-5.2 (max)Terminus-295GLM-5.3-Flash (low)Terminus-295GLM-5.3 (low)Terminus-295GLM-5.3 (max)Claude Code95GLM-5.3-Flash (max)Terminus-284GLM-5.3 (max)Terminus-284GLM-5.2 (Non-reasoning)Terminus-274GLM-4.7-Flash (Reasoning)Terminus-237GLM-4.7-Flash (Reasoning)Claude Code26
Data table: GeoAnalystBench
GeoAnalystBench
ModelDeveloperGeoAnalystBench
GLM-5.2 (high), Terminus-2Z.ai100
GLM-5.3-Flash (max), Claude CodeZ.ai100
DeepSeek V4 Pro (max), Terminus-2DeepSeek100
Claude Opus 5.5 (high), Claude CodeAnthropic100
GLM-5.2 (max), Terminus-2Z.ai95
GLM-5.3-Flash (low), Terminus-2Z.ai95
GLM-5.3 (low), Terminus-2Z.ai95
GLM-5.3 (max), Claude CodeZ.ai95
GLM-5.3-Flash (max), Terminus-2Z.ai84
GLM-5.3 (max), Terminus-2Z.ai84
GLM-5.2 (Non-reasoning), Terminus-2Z.ai74
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai37
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai26

OGD4All

Not in an index: questions on a city's open government data · 0 to 100 · Higher is better

OGD4AllScores 0 to 100: Claude Opus 4.8 (high), Claude Code: 97; GLM-5.3 (max), Claude Code: 96; Qwen 3.8 27B (xhigh), Qwen Code: 95; GLM-5.3-Flash (max), Claude Code: 930255075100Claude Opus 4.8 (high)Claude Code97GLM-5.3 (max)Claude Code96Qwen 3.8 27B (xhigh)Qwen Code95GLM-5.3-Flash (max)Claude Code93
Data table: OGD4All
OGD4All
ModelDeveloperOGD4All
Claude Opus 4.8 (high), Claude CodeAnthropic97
GLM-5.3 (max), Claude CodeZ.ai96
Qwen 3.8 27B (xhigh), Qwen CodeAlibaba (Qwen)95
GLM-5.3-Flash (max), Claude CodeZ.ai93

GeoNatureAgent

Not in an index: environmental raster analysis over named regions · 0 to 100 · Higher is better

GeoNatureAgentScores 0 to 100: GLM-5.3-Flash (max), Claude Code: 100; GLM-5.3 (max), Claude Code: 100; Qwen 3.8 27B (xhigh), Terminus-2: 100; Step 5 Preview (Reasoning), Terminus-2: 1000255075100GLM-5.3-Flash (max)Claude Code100GLM-5.3 (max)Claude Code100Qwen 3.8 27B (xhigh)Terminus-2100Step 5 Preview (Reasoning)Terminus-2100
Data table: GeoNatureAgent
GeoNatureAgent
ModelDeveloperGeoNatureAgent
GLM-5.3-Flash (max), Claude CodeZ.ai100
GLM-5.3 (max), Claude CodeZ.ai100
Qwen 3.8 27B (xhigh), Terminus-2Alibaba (Qwen)100
Step 5 Preview (Reasoning), Terminus-2StepFun100

MultiGlobeQA

Not in an index: spatial reasoning over facts about places worldwide · 0 to 100 · Higher is better

MultiGlobeQAScores 0 to 100: Step 5 Preview (Reasoning), Terminus-2: 83; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 82; GLM-5.3 (max), Claude Code: 74; GLM-5.3-Flash (max), Claude Code: 730255075100Step 5 Preview (Reasoning)Terminus-283Nemotron 3.5 Super VL 120B (Reasoning)Terminus-282GLM-5.3 (max)Claude Code74GLM-5.3-Flash (max)Claude Code73
Data table: MultiGlobeQA
MultiGlobeQA
ModelDeveloperMultiGlobeQA
Step 5 Preview (Reasoning), Terminus-2StepFun83
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA82
GLM-5.3 (max), Claude CodeZ.ai74
GLM-5.3-Flash (max), Claude CodeZ.ai73

GeoAgentEval

Not in an index: building, auditing and repairing a PostGIS database · 0 to 100 · Higher is better

GeoAgentEvalScores 0 to 100: Claude Opus 5.5 (medium), Claude Code: 100; GLM-5.3-Flash (max), Claude Code: 96; Step 5 Preview (Reasoning), Terminus-2: 95; GLM-5.3 (max), Claude Code: 880255075100Claude Opus 5.5 (medium)Claude Code100GLM-5.3-Flash (max)Claude Code96Step 5 Preview (Reasoning)Terminus-295GLM-5.3 (max)Claude Code88
Data table: GeoAgentEval
GeoAgentEval
ModelDeveloperGeoAgentEval
Claude Opus 5.5 (medium), Claude CodeAnthropic100
GLM-5.3-Flash (max), Claude CodeZ.ai96
Step 5 Preview (Reasoning), Terminus-2StepFun95
GLM-5.3 (max), Claude CodeZ.ai88

Vision

ThinkGeo

Vision agents: agents that look at images and use code · 0 to 100 · Higher is better

ThinkGeoScores 0 to 100: Claude Opus 5.5 (xhigh), Claude Code: 96; GPT-6.1 Sol (xhigh), Codex: 88; GLM-5.3-Flash (max), Claude Code: 74025507510096Claude Opus5.5 (xhigh)Claude Code88GPT-6.1 Sol(xhigh)Codex74GLM-5.3-Flash(max)Claude Code
Data table: ThinkGeo
ThinkGeo
ModelDeveloperThinkGeo
Claude Opus 5.5 (xhigh), Claude CodeAnthropic96
GPT-6.1 Sol (xhigh), CodexOpenAI88
GLM-5.3-Flash (max), Claude CodeZ.ai74

OpenEarthAgent

Vision agents: agents that look at images and use code · 0 to 100 · Higher is better

OpenEarthAgentScores 0 to 100: Claude Opus 5.5 (medium), Claude Code: 100; GPT-6.1 Sol (medium), Codex: 100; GLM-5.3-Flash (max), Claude Code: 880255075100100Claude Opus5.5 (medium)Claude Code100GPT-6.1 Sol(medium)Codex88GLM-5.3-Flash(max)Claude Code
Data table: OpenEarthAgent
OpenEarthAgent
ModelDeveloperOpenEarthAgent
Claude Opus 5.5 (medium), Claude CodeAnthropic100
GPT-6.1 Sol (medium), CodexOpenAI100
GLM-5.3-Flash (max), Claude CodeZ.ai88

Earth-Bench V2 images

Questions on images, no tools · 0 to 100 · Higher is better

Earth-Bench V2 imagesScores 0 to 100: GPT-5.6 Sol (high): 76; Claude Opus 5.5 (high): 74; GLM-5.3-Flash (low): 56; GLM-5.3-Flash (max): 530255075100GPT-5.6 Sol (high)76Claude Opus 5.5 (high)74GLM-5.3-Flash (low)56GLM-5.3-Flash (max)53
Data table: Earth-Bench V2 images
Earth-Bench V2 images
ModelDeveloperEarth-Bench V2 images
GPT-5.6 Sol (high)OpenAI76
Claude Opus 5.5 (high)Anthropic74
GLM-5.3-Flash (low)Z.ai56
GLM-5.3-Flash (max)Z.ai53

ThinkGeo images

Questions on images, no tools · 0 to 100 · Higher is better

ThinkGeo imagesScores 0 to 100: Claude Sonnet 5.5 (high): 50; GPT-6 Luna (high): 43; GLM-5.3-Flash (low): 36; GLM-5.3-Flash (max): 350255075100Claude Sonnet 5.5 (high)50GPT-6 Luna (high)43GLM-5.3-Flash (low)36GLM-5.3-Flash (max)35
Data table: ThinkGeo images
ThinkGeo images
ModelDeveloperThinkGeo images
Claude Sonnet 5.5 (high)Anthropic50
GPT-6 Luna (high)OpenAI43
GLM-5.3-Flash (low)Z.ai36
GLM-5.3-Flash (max)Z.ai35

Knowledge

Geospatial knowledge

Questions on geospatial knowledge · 0 to 100 · Higher is better

Geospatial knowledgeScores 0 to 100: Claude Sonnet 4.6 (high): 98; Gemini 3.8 Flash (low): 94; GLM-5.3 (max): 92; GLM-5.3-Flash (max): 92; GLM-5.2 (high): 91; GLM-5.3 (low): 91; GLM-5.2 (Non-reasoning): 89; GLM-5.2 (max): 89; GLM-4.7-Flash (Reasoning): 86; GLM-5.3-Flash (low): 850255075100Claude Sonnet 4.6 (high)98Gemini 3.8 Flash (low)94GLM-5.3 (max)92GLM-5.3-Flash (max)92GLM-5.2 (high)91GLM-5.3 (low)91GLM-5.2 (Non-reasoning)89GLM-5.2 (max)89GLM-4.7-Flash (Reasoning)86GLM-5.3-Flash (low)85
Data table: Geospatial knowledge
Geospatial knowledge
ModelDeveloperGeospatial knowledge
Claude Sonnet 4.6 (high)Anthropic98
Gemini 3.8 Flash (low)Google94
GLM-5.3 (max)Z.ai92
GLM-5.3-Flash (max)Z.ai92
GLM-5.2 (high)Z.ai91
GLM-5.3 (low)Z.ai91
GLM-5.2 (Non-reasoning)Z.ai89
GLM-5.2 (max)Z.ai89
GLM-4.7-Flash (Reasoning)Z.ai86
GLM-5.3-Flash (low)Z.ai85

Cost

Cost per task

Average cost per agent task (USD) · Lower is better

Cost per taskCost per task in US dollars: GLM-4.7-Flash (Reasoning), Terminus-2: $0.037; GLM-5.3-Flash (low), Terminus-2: $0.048; GPT-6.1 Sol (xhigh), Codex: $0.085; GLM-5.3-Flash (max), Claude Code: $0.093; GLM-5.3-Flash (max), Terminus-2: $0.094; GLM-5.2 (Non-reasoning), Terminus-2: $0.099; GLM-5.2 (high), Terminus-2: $0.22; Claude Opus 5.5 (xhigh), Claude Code: $0.25; GLM-4.7-Flash (Reasoning), Claude Code: $0.27; GLM-5.3 (low), Terminus-2: $0.27; GLM-5.2 (max), Terminus-2: $0.32; GLM-5.2 (max), Claude Code: $0.34; GLM-5.3 (max), Terminus-2: $0.38; GLM-5.3 (max), Claude Code: $0.48$0$0.15$0.30$0.45$0.60GLM-4.7-Flash (Reasoning)Terminus-2$0.037GLM-5.3-Flash (low)Terminus-2$0.048GPT-6.1 Sol (xhigh)Codex$0.085GLM-5.3-Flash (max)Claude Code$0.093GLM-5.3-Flash (max)Terminus-2$0.094GLM-5.2 (Non-reasoning)Terminus-2$0.099GLM-5.2 (high)Terminus-2$0.22Claude Opus 5.5 (xhigh)Claude Code$0.25GLM-4.7-Flash (Reasoning)Claude Code$0.27GLM-5.3 (low)Terminus-2$0.27GLM-5.2 (max)Terminus-2$0.32GLM-5.2 (max)Claude Code$0.34GLM-5.3 (max)Terminus-2$0.38GLM-5.3 (max)Claude Code$0.48
How to read this chart

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How to read this chart. Lower is better. Costs are pay-per-token list prices, not any discount or credit a customer might have.

Bars. Every Z.ai entry with a cost, highlighted, and the top-ranked entries from other labs on the Geospatial Agent Index, faded, for context. Provisional: not yet ranked.

Data table: Cost per task
Cost per task
ModelDeveloperCost per task
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai$0.037
GLM-5.3-Flash (low), Terminus-2Z.ai$0.048
GPT-6.1 Sol (xhigh), CodexOpenAI$0.085
GLM-5.3-Flash (max), Claude CodeZ.ai$0.093
GLM-5.3-Flash (max), Terminus-2Z.ai$0.094
GLM-5.2 (Non-reasoning), Terminus-2Z.ai$0.099
GLM-5.2 (high), Terminus-2Z.ai$0.22
Claude Opus 5.5 (xhigh), Claude CodeAnthropic$0.25
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai$0.27
GLM-5.3 (low), Terminus-2Z.ai$0.27
GLM-5.2 (max), Terminus-2Z.ai$0.32
GLM-5.2 (max), Claude CodeZ.ai$0.34
GLM-5.3 (max), Terminus-2Z.ai$0.38
GLM-5.3 (max), Claude CodeZ.ai$0.48

Geospatial Agent Index vs. cost per task

Geospatial Agent Index, Text only; cost per task (USD) · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
Geospatial Agent Index vs. cost per taskGeospatial Agent Index (0 to 100) against cost per task in US dollars, log scale: GLM-4.7-Flash (Reasoning), Terminus-2: 43.17, $0.037; GLM-5.2 (high), Terminus-2: 70.82, $0.22; GLM-5.2 (Non-reasoning), Terminus-2: 63.34, $0.099; GLM-5.2 (max), Terminus-2: 71.73, $0.32; GLM-5.3-Flash (low), Terminus-2: 61.76, $0.048; GLM-5.3-Flash (max), Terminus-2: 62.51, $0.094; GLM-5.3 (low), Terminus-2: 65.65, $0.27; GLM-5.3 (max), Terminus-2: 70.45, $0.38; GLM-4.7-Flash (Reasoning), Claude Code: 41.27, $0.27; GLM-5.2 (max), Claude Code: 73.32, $0.34; GLM-5.3-Flash (max), Claude Code: 71.56, $0.093; GLM-5.3 (max), Claude Code: 73.4, $0.48; Claude Opus 5.5 (xhigh), Claude Code: 84.65, $0.25; GPT-6.1 Sol (xhigh), Codex: 81.82, $0.085; Kimi K3 (Reasoning), Terminus-2: 71.99, $0.23Most attractive quadrant30405060708090$0.02$0.05$0.10$0.20$0.50$1.00Cost per task (USD, log scale)Geospatial Agent IndexGLM-5.3-Flash (max), Claude CodeGLM-5.2 (Non-reasoning), Terminus-2GLM-5.3-Flash (low), Terminus-2GLM-4.7-Flash (Reasoning), Terminus-2GLM-5.3 (max), Claude CodeGLM-5.2 (max), Claude CodeGLM-5.2 (max), Terminus-2GLM-5.2 (high), Terminus-2GLM-5.3 (max), Terminus-2GLM-5.3 (low), Terminus-2GLM-5.3-Flash (max), Terminus-2GLM-4.7-Flash (Reasoning), Claude CodeClaude Opus 5.5 (xhigh), Claude CodeGPT-6.1 Sol (xhigh), CodexKimi K3 (Reasoning), Terminus-2
How to read this chart

Why a log scale. Costs and times differ by a factor of ten or more between models, so equal steps on the axis are equal ratios.

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How to read this chart. Each point is one entry. Further left means a lower average cost per task (log scale); higher means a higher Geospatial Agent 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 that moves with the axes when you change the models shown, not a fixed bar. The dotted Pareto line joins the shown models that no cheaper shown model beats on score.

Data table: Geospatial Agent Index vs. cost per task
Geospatial Agent Index vs. cost per task
ModelGeospatial Agent IndexCost per task (USD)On the Pareto line
GLM-4.7-Flash (Reasoning), Terminus-243$0.037Yes
GLM-5.2 (high), Terminus-271$0.22No
GLM-5.2 (Non-reasoning), Terminus-263$0.099No
GLM-5.2 (max), Terminus-272$0.32No
GLM-5.3-Flash (low), Terminus-262$0.048Yes
GLM-5.3-Flash (max), Terminus-263$0.094No
GLM-5.3 (low), Terminus-266$0.27No
GLM-5.3 (max), Terminus-270$0.38No
GLM-4.7-Flash (Reasoning), Claude Code41$0.27No
GLM-5.2 (max), Claude Code73$0.34No
GLM-5.3-Flash (max), Claude Code72$0.093No
GLM-5.3 (max), Claude Code73$0.48No
Claude Opus 5.5 (xhigh), Claude Code85$0.25Yes
GPT-6.1 Sol (xhigh), Codex82$0.085Yes
Kimi K3 (Reasoning), Terminus-272$0.23No

Time

Time per task

Average minutes per agent task, at 100 output tokens/s · Lower is better

Time per taskTime per task: GPT-6.1 Sol (xhigh), Codex: 0.8 min; Claude Opus 5.5 (xhigh), Claude Code: 1.4 min; GLM-5.2 (Non-reasoning), Terminus-2: 1.5 min; GLM-5.3 (low), Terminus-2: 2.7 min; GLM-4.7-Flash (Reasoning), Claude Code: 3.8 min; GLM-5.3 (max), Terminus-2: 4.2 min; GLM-5.3-Flash (low), Terminus-2: 4.2 min; GLM-5.3 (max), Claude Code: 4.3 min; GLM-5.2 (high), Terminus-2: 4.7 min; GLM-4.7-Flash (Reasoning), Terminus-2: 5.4 min; GLM-5.2 (max), Terminus-2: 6.1 min; GLM-5.2 (max), Claude Code: 6.3 min; GLM-5.3-Flash (max), Claude Code: 6.5 min; GLM-5.3-Flash (max), Terminus-2: 8.5 min02.557.510GPT-6.1 Sol (xhigh)Codex0.8 minClaude Opus 5.5 (xhigh)Claude Code1.4 minGLM-5.2 (Non-reasoning)Terminus-21.5 minGLM-5.3 (low)Terminus-22.7 minGLM-4.7-Flash (Reasoning)Claude Code3.8 minGLM-5.3 (max)Terminus-24.2 minGLM-5.3-Flash (low)Terminus-24.2 minGLM-5.3 (max)Claude Code4.3 minGLM-5.2 (high)Terminus-24.7 minGLM-4.7-Flash (Reasoning)Terminus-25.4 minGLM-5.2 (max)Terminus-26.1 minGLM-5.2 (max)Claude Code6.3 minGLM-5.3-Flash (max)Claude Code6.5 minGLM-5.3-Flash (max)Terminus-28.5 min
How to read this chart

What this metric means. 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.

Bars. Every Z.ai entry with a cost, highlighted, and the top-ranked entries from other labs on the Geospatial Agent Index, faded, for context. Provisional: not yet ranked.

Data table: Time per task
Time per task
ModelDeveloperTime per task (at 100 output tokens/s)
GPT-6.1 Sol (xhigh), CodexOpenAI0.8 min
Claude Opus 5.5 (xhigh), Claude CodeAnthropic1.4 min
GLM-5.2 (Non-reasoning), Terminus-2Z.ai1.5 min
GLM-5.3 (low), Terminus-2Z.ai2.7 min
GLM-4.7-Flash (Reasoning), Claude CodeZ.ai3.8 min
GLM-5.3 (max), Terminus-2Z.ai4.2 min
GLM-5.3-Flash (low), Terminus-2Z.ai4.2 min
GLM-5.3 (max), Claude CodeZ.ai4.3 min
GLM-5.2 (high), Terminus-2Z.ai4.7 min
GLM-4.7-Flash (Reasoning), Terminus-2Z.ai5.4 min
GLM-5.2 (max), Terminus-2Z.ai6.1 min
GLM-5.2 (max), Claude CodeZ.ai6.3 min
GLM-5.3-Flash (max), Claude CodeZ.ai6.5 min
GLM-5.3-Flash (max), Terminus-2Z.ai8.5 min

Geospatial Agent Index vs. time per task

Geospatial Agent Index, Text only; time per task (minutes) · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
Geospatial Agent Index vs. time per taskGeospatial Agent Index (0 to 100) against time per task (minutes), log scale: GLM-4.7-Flash (Reasoning), Terminus-2: 43.17, 5.4 min; GLM-5.2 (high), Terminus-2: 70.82, 4.7 min; GLM-5.2 (Non-reasoning), Terminus-2: 63.34, 1.5 min; GLM-5.2 (max), Terminus-2: 71.73, 6.1 min; GLM-5.3-Flash (low), Terminus-2: 61.76, 4.2 min; GLM-5.3-Flash (max), Terminus-2: 62.51, 8.5 min; GLM-5.3 (low), Terminus-2: 65.65, 2.7 min; GLM-5.3 (max), Terminus-2: 70.45, 4.2 min; GLM-4.7-Flash (Reasoning), Claude Code: 41.27, 3.8 min; GLM-5.2 (max), Claude Code: 73.32, 6.3 min; GLM-5.3-Flash (max), Claude Code: 71.56, 6.5 min; GLM-5.3 (max), Claude Code: 73.4, 4.3 min; Claude Opus 5.5 (xhigh), Claude Code: 84.65, 1.4 min; GPT-6.1 Sol (xhigh), Codex: 81.82, 0.8 min; Kimi K3 (Reasoning), Terminus-2: 71.99, 2.7 minMost attractive quadrant304050607080900.5 min1 min2 min5 min10 minTime per task (minutes, log scale)Geospatial Agent IndexGLM-5.2 (Non-reasoning), Terminus-2GLM-5.3 (max), Claude CodeGLM-5.2 (max), Claude CodeGLM-5.2 (max), Terminus-2GLM-5.3-Flash (max), Claude CodeGLM-5.2 (high), Terminus-2GLM-5.3 (max), Terminus-2GLM-5.3 (low), Terminus-2GLM-5.3-Flash (max), Terminus-2GLM-5.3-Flash (low), Terminus-2GLM-4.7-Flash (Reasoning), Terminus-2GLM-4.7-Flash (Reasoning), Claude CodeClaude Opus 5.5 (xhigh), Claude CodeGPT-6.1 Sol (xhigh), CodexKimi K3 (Reasoning), Terminus-2
How to read this chart

Why a log scale. Costs and times differ by a factor of ten or more between models, so equal steps on the axis are equal ratios.

What this metric means. 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.

How to read this chart. Each point is one entry. Further left means a lower average time per task (log scale); higher means a higher Geospatial Agent 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 that moves with the axes when you change the models shown, not a fixed bar. The dotted Pareto line joins the shown models that no faster shown model beats on score.

Data table: Geospatial Agent Index vs. time per task
Geospatial Agent Index vs. time per task
ModelGeospatial Agent IndexTime per task (minutes)On the Pareto line
GLM-4.7-Flash (Reasoning), Terminus-2435.4 minNo
GLM-5.2 (high), Terminus-2714.7 minNo
GLM-5.2 (Non-reasoning), Terminus-2631.5 minNo
GLM-5.2 (max), Terminus-2726.1 minNo
GLM-5.3-Flash (low), Terminus-2624.2 minNo
GLM-5.3-Flash (max), Terminus-2638.5 minNo
GLM-5.3 (low), Terminus-2662.7 minNo
GLM-5.3 (max), Terminus-2704.2 minNo
GLM-4.7-Flash (Reasoning), Claude Code413.8 minNo
GLM-5.2 (max), Claude Code736.3 minNo
GLM-5.3-Flash (max), Claude Code726.5 minNo
GLM-5.3 (max), Claude Code734.3 minNo
Claude Opus 5.5 (xhigh), Claude Code851.4 minYes
GPT-6.1 Sol (xhigh), Codex820.8 minYes
Kimi K3 (Reasoning), Terminus-2722.7 minNo

Every benchmark

Score by benchmark

Share answered correctly, 0 to 100 · Darker is better; hatched: provisional, evaluation not finished

Data table: Score by benchmark
Every benchmark
ModelStatusEarth-Bench V2GeoBenchXGS-QA2GeoAgentBenchGeoAnalystBenchOGD4AllGeoNatureAgentMultiGlobeQAGeoAgentEvalThinkGeoOpenEarthAgentEarth-Bench V2 imagesThinkGeo imagesGeospatial knowledge
GLM-5.3 (max), Claude CodeRanked7465819895961007488No dataNo dataNo dataNo data92
GLM-5.2 (max), Claude CodeRanked716683No dataNo dataNo dataNo dataNo dataNo dataNo dataNo dataNo dataNo data89
GLM-5.2 (max), Terminus-2Ranked6867809495No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data89
GLM-5.3-Flash (max), Claude CodeRanked697076981009310073967488533592
GLM-5.2 (high), Terminus-2Ranked68677894100No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data91
GLM-5.3 (max), Terminus-2Ranked6465839484No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data92
GLM-5.3 (low), Terminus-2Ranked5760809295No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data91
GLM-5.2 (Non-reasoning), Terminus-2Ranked5860729274No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data89
GLM-5.3-Flash (max), Terminus-2Ranked5157808484No dataNo dataNo dataNo dataNo dataNo data533592
GLM-5.3-Flash (low), Terminus-2Ranked5357769495No dataNo dataNo dataNo dataNo dataNo data563685
GLM-4.7-Flash (Reasoning), Terminus-2Ranked4442434237No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data86
GLM-4.7-Flash (Reasoning), Claude CodeRanked3839483026No dataNo dataNo dataNo dataNo dataNo dataNo dataNo data86

Frequently asked questions

What Z.ai models do you benchmark?
We benchmark 4 Z.ai models: GLM-5.3, GLM-5.2, GLM-5.3-Flash and GLM-4.7-Flash.
Which is the most intelligent Z.ai model for geospatial work?
GLM-5.2 (high), with 78.6 on the Geospatial Intelligence Index (Text only), #27 of 61.
What is the cheapest Z.ai model per task?
GLM-4.7-Flash (Reasoning), Terminus-2, at $0.037 per agent task at list prices, #25 of 74 on the Geospatial Agent Index.
What is the fastest Z.ai model?
GLM-5.2 (Non-reasoning), Terminus-2, at 1.5 min per agent task at 100 output tokens per second, #32 of 74.
Does Z.ai offer open-weights models?
Yes: GLM-5.3, GLM-5.2, GLM-5.3-Flash and GLM-4.7-Flash. Their weights are published, so you can run them on your own hardware.
Which Z.ai models can read satellite imagery?
GLM-5.3-Flash takes images as input. The best on Vision is GLM-5.3-Flash (low), with 45.8.
How do Z.ai models do on Earth observation, GIS and geospatial knowledge?
On the Geospatial Agent Index, the best is GLM-5.3 (max), Claude Code, with 73.4. On Vision, GLM-5.3-Flash (low) scores 45.8. On Geospatial knowledge, GLM-5.3 (max) scores 91.8.
Which Z.ai model is best for agent work, and in which harness?
GLM-5.3 (max), Claude Code, with 73.4 on the Geospatial Agent Index, #18 of 74, in the Claude Code harness.
How does Z.ai's best model compare with the overall leader?
GLM-5.2 (high) scores 78.6 on the Geospatial Intelligence Index (Text only), 10.0 points behind the leader, Anthropic's Claude Opus 5.5 (xhigh), with 88.7.
Are any Z.ai models still being benchmarked?
No. Every Z.ai model listed has finished its benchmark and is ranked.
How should I choose between Z.ai models for geospatial work?
Start from the job. For the strongest answers, GLM-5.2 (high); for agent work with data and code, GLM-5.3 (max), Claude Code; for the lowest cost, GLM-4.7-Flash (Reasoning), Terminus-2; for reading imagery, GLM-5.3-Flash (low); to run it yourself, GLM-5.3. Each model page shows its scores on every evaluation.