Axis Spatial v2 - Earth-Bench V2: leaderboard

Spatial Agent Index v2, separate from methodology1.0 observations. One attempt per task, 60turns and7200seconds; native reasoning/harness defaults, output cap65536except native ClaudeCode/Codex. Provider calls retry up to9times; persistent failed trajectories rerun, at most1%after3reruns omitted. Refusals score0. A timeout with over half time waiting on provider reruns. Unattempted/unjudged maps omitted; passing attempts awaiting trace review count as passes. Trace failure is limited to obtaining answer/check content or external answers. Claude Sonnet5.5 high judges maps and traces, including its own family. Nominal Vision pool128built minus47excluded=81; source history adds22ThinkGeo tasks, while an older106-42=64paragraph is stale. Native per-entry scored denominators remain unreported. Same-model/effort Stirrup Vision results may supply Terminus2 entries. At least75%view completion qualifies a source index entry; default charts select model families and do not define our retained roster. Metric: Pass@1 (%). Source: https://www.axisspatial.com/arena/agent-index

Metric: Pass@1 (%). Source: www.axisspatial.com. Saturation forecast: Rough model projection: around 2026. 50 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 5.5(Max) (Axis Spatial run: native=claude-code-claude-opus-5-5-max, cfg=06d59074e2e49c42)80.17#7
2DeepSeek V4.1 Flash(Max) (Axis Spatial run: native=claude-code-deepseek-v4-1-flash-max, cfg=83a4668e8c55fd0f)76.92#69
3GLM-5.3(Max) (Axis Spatial run: native=claude-code-glm-5-3-max-default, cfg=a70ebf1edca00a58)76.92#73
4GPT-6 Sol(Medium) (Axis Spatial run: native=codex-gpt-6-sol-medium, cfg=9cfd2a064105bb74)76.71#29
5Kimi K3(Reasoning) (Axis Spatial run: native=kimi-code-kimi-k3-always-on, cfg=fb91857695683970)75.21#59
6DeepSeek V4 Pro(High) (Axis Spatial run: native=codex-deepseek-v4-pro-high, cfg=21c081d31d0b1e76)75#198
7Claude Opus 5(High) (Axis Spatial run: native=claude-code-claude-opus-5-high, cfg=6285e6a048aeccf0)74.36#23
8Claude Fable 5.1(Medium) (Axis Spatial run: native=claude-code-claude-fable-5-1-medium, cfg=d02f729401d534f3)72.65#17
9DeepSeek V4 Flash(High) (Axis Spatial run: native=deepseek-v4-flash-high, cfg=2924aebe5a92c7fd)72.65#265
10Kimi K2.7 Code(Reasoning) (Axis Spatial run: native=kimi-code-kimi-k2-7-code-always-on, cfg=e1a8c32ea8eb295e)72.65#199
11GPT-6.1 Sol(xHigh) (Axis Spatial run: native=codex-gpt-6-1-sol-xhigh, cfg=006589addbd990d4)71.79#12
12Claude Haiku 5.5(High) (Axis Spatial run: native=claude-code-claude-haiku-5-5-high, cfg=853ae92b1f80a944)71.79#76
13GLM-5.2(Max) (Axis Spatial run: native=claude-code-glm-5-2-max, cfg=eb741e1eb6bdfc0d)71.3#152
14Claude Sonnet 5.5(High) (Axis Spatial run: native=claude-code-claude-sonnet-5-5-high, cfg=bf92f603d195d150)70.94#21
15Claude Opus 4.8(High) (Axis Spatial run: native=claude-code-claude-opus-4-8-high, cfg=5dcc0d54f0927527)70.94#62

Interactive version: theaggregate.ai/benchmark?slug=axis-spatial-v2-earth-bench-v2 · How It Works · Data refreshed daily, snapshot 2026-11-17.