// by MaximVLgraphatlas.techgeobrowser.io
{{ screenTitle }}
{{ screenSub }}
GraphAtlasGeoGitHub
Search agents, capabilities, protocols… /
Filters
MODEL PRICE/PERFORMANCE
What does it cost to solve a task?
X: avg cost per configuration (log $) · Y: avg score across benchmark runs · size: evaluated runs
Score is averaged across whatever benchmarks each model happened to run, so cross-model comparisons are directional, not apples-to-apples.
100% 0% {{ modelLandscape.costMinLabel }} {{ modelLandscape.costMaxLabel }} {{ sg.label }} {{ cg.label }}
{{ lg.label }} Pareto front
Best score/cost tradeoff: {{ modelLandscape.topModel.name }} · {{ modelLandscape.topModel.cost }} · {{ modelLandscape.topModel.score }}
Agent landscape
X: task execution depth · Y: ops/control signals · size: GitHub activity
DERIVED VIEW
autonomy / execution depth
ops & control signals
Coding Browser Orchestration Memory opacity = record coverage confidence proxy
Capability coverage heatmap
Cluster density across core capabilities
{{ cap.abbr }} {{ row.name }}
Record coverage
96%
133 records · source-of-truth JSONweak ◦ → ◦ strong
TASK FIT
Which agent fits this scenario?
Top 5 by task-fit score. Expand a row for the signals behind it.
{{ s.label }}
{{ r.chev }} #{{ r.rank }}
{{ r.name }}
{{ r.vendor }}
{{ r.score }}
WHY IT SCORES HERE
• {{ x.l }}
TRADE-OFFS
• {{ x.l }}
CAPABILITY SIGNALS
{{ x.l }}
+ compare
SCORE WEIGHTING
{{ w.label }} {{ w.pct }}
DATASET STATISTICS
{{ m.label }} {{ m.value }}
FILTERS · AND {{ lens.label }} · {{ lens.count }} clear
{{ matrixFilterSummary }} density low → high
Agent {{ sortArrowName }}
Capabilities
{{ c.abbr }}
Protocols
Model
Runtimes
Memory
License
Docs {{ sortArrowDocs }}
GitHub
Maint
VERIFIED {{ sortArrowAtg }}
{{ row.name }}geo ↗
{{ row.vendor }}{{ row.confBadge.label }}
{{ row.capCount }}
{{ p.l }}
{{ row.model }}
{{ r.l }}
{{ row.memory.l }}
{{ row.lic.label }}
{{ row.docs.label }}
{{ row.stars }}
{{ row.maint.label }}
{{ row.verified.v }}
SELECT 2–3 AGENTS {{ compareCount }} selected
{{ diffToggleLabel }}
GROUP {{ g.label }} · {{ g.count }}
{{ p.name }}
EDGE EVIDENCE
{{ evidencePanel.agent }} → {{ evidencePanel.capability }}
{{ evidencePanel.confidence }}{{ evidencePanel.method }}{{ evidencePanel.date }}
close
{{ evidencePanel.excerpt }}
open source · {{ evidencePanel.host }}
{{ c.name }}
{{ c.vendor }}
{{ c.atg.v }}
COVERAGE
{{ c.cluster }} {{ c.confBadge.label }} Geo card ↗
EXPLAINABLE SCORES
{{ s.label }} {{ s.v }}
CAPABILITIES · CLICK FOR EVIDENCE
{{ x.l }}{{ x.tag }}
PROTOCOLS
{{ x.l }}
MODEL
{{ c.model }}
{{ c.observedLabel }}
RUNTIMES
{{ c.runtimes }}
MEMORY
{{ c.memory.l }}
LICENSE
{{ c.lic.label }}
DOCS
{{ c.docs.label }}
EVIDENCE
{{ c.ev.label }}
BENCHMARK EVIDENCE
{{ c.bench }}
{{ b.key }}{{ b.score }}
{{ b.model }}
{{ b.status }}
BENCHMARKS · SCORE × COST
no eval-run records for this agent
{{ row.benchmark }} {{ row.model }} {{ row.score }} {{ row.cost }} {{ row.sourceLabel }} {{ row.sourceLabel }}
no priced runs recorded for this agent — market context only
{{ p.label }} ({{ p.count }})
{{ c.benchEval.active.costMinLabel }} {{ c.benchEval.active.costMaxLabel }} 100% 0%
cost (log $) · {{ c.benchEval.activeLabel }} cyan = this agent · gray = other agents · gold = Pareto front
ACTIVE RISKS · RULE-DERIVED
{{ x.l }}
SURFACE
{{ c.riskSurfaceCount }}
MITIGATED
{{ c.riskMitigatedCount }}
Exposure minus mapped controls; not a security audit.
WHY IT FITS
• {{ x.l }}
TRADE-OFFS
• {{ x.l }}
EXTERNAL REFERENCE SIGNALS
Hands-On AI Engineering comparison
{{ externalSignals.summary }}
open source repo
DIRECT AGENT ADDS
{{ externalSignals.directAgentAdds }}
FRAMEWORK MATCHES
{{ externalSignals.matchCount }}
REFERENCE APPS
{{ externalSignals.referenceCount }}
NEW LAYERS
{{ externalSignals.layerCount }}
MISSING FRAMEWORK CANDIDATE
{{ m.name }}
{{ m.type }}
{{ m.reason }}
{{ m.status }}
NEW ENTITY LAYERS
{{ x.type }}
{{ x.why }}
{{ x.examples }}
CURATED REFERENCE IMPLEMENTATIONS
{{ x.name }}
built with: {{ x.builtWith }}
MCP: {{ x.mcpServers }}
{{ x.note }}
Cluster lenses 6 LENSES · GROUPED BY BEHAVIOUR
{{ cl.name }}
{{ cl.count }}

{{ cl.desc }}

{{ m.l }} {{ cl.more }}
AVG ATG
{{ cl.avg }}
Coverage gaps WHERE THE ATLAS IS THIN
{{ g.count }}
{{ g.name }}

{{ g.desc }}

{{ m.l }}
{{ h.value }}
{{ h.detail }}
Cheapest solve vs. adoption
Each dot is one task · X: cheapest successful run, log $ · Y: priced passing runs per task · dot size: price spread · hover for details
{{ tk.label }}
{{ taskPrices.maxSolvers }} passing runs
Task price spread
{{ taskPrices.taskCount }} of {{ taskPrices.totalTasks }} Terminal-Bench 2.0 tasks have at least one solved, priced run ({{ taskPrices.unpricedTasks }} were solved only by configs that don't report cost) · click a row for the 20 cheapest successful runs
Rows in the expanded list are individual successful runs, not deduped configs — the same agent × model pair can appear several times at different prices: every trial takes its own trajectory (steps, retries, tokens), so cost varies run to run even with the agent and model fixed.
SORT {{ s.label }}
Task
Price range
Passing runs
Success
Min cost
Median
Spread
{{ row.chev }} {{ row.task_name }}
{{ row.solvers }}
{{ row.successLabel }}
{{ row.minCostLabel }}
{{ row.medianCostLabel }}
{{ row.spreadLabel }}
CHEAPEST SUCCESSFUL RUNS
#{{ c.rank }} {{ c.label }} {{ c.costLabel }}
RELATION NETWORK · CENTERED ON
{{ graph.center.name }}
TYPED EDGES
{{ lg.l }}
{{ n.label }}
{{ graph.center.name }}
click a similarity node to re-center →
{{ graph.center.name }}
MAINTAINER
{{ graph.meta.vendor }}
CLUSTER
{{ graph.meta.cluster }}
MODEL
{{ graph.meta.model }}
LICENSE
{{ graph.meta.lic }}
EXPLAINABLE SCORE CARDS
{{ s.label }}{{ s.v }}
RE-CENTER GRAPH
{{ p.name }}
Mobile adaptation SAME ATLAS · COMPRESSED FOR TOUCH

The dense matrix collapses into scannable agent cards; comparison switches to a vertical dual-bar; score pills and heatmap cells keep their meaning. Filters move into a bottom sheet.

9:41
ATG.atlas
TAXONOMY ATLAS
AI Agents
Taxonomy Graph
AGENTS
17
RELATIONS
312
CAPABILITIES
18
RISKS
24
TOP BY ATG
Claude Code
91
LangGraph
86
Cursor Agent
84
OVERVIEW
MATRIX
COMPARE
GRAPH
Overview
9:41
Agent Matrix
Search agents…
All · 17MCPOpen
Claude Code
Anthropic
91
MCP
OpenHands
All Hands AI
82
MIT
Browser Use
Browser Use
73
MIT
Matrix → agent cards
9:41
Compare
2 of 5 agents
Claude CodeCursor
91
CLAUDE CODE
84
CURSOR
Coding fit95 · 90
Browser fit40 · 35
Ops/control82 · 76
Record coverage92 · 72
proprietaryproprietary
Compare → vertical dual-bar