Axis Spatial Spatial Intelligence Index

Search modelsMenu

DLPU phase unwrapping benchmark leaderboard

Unwrapping simulated radar interferograms, from DLPU phase unwrapping. A standalone evaluation: its scores are shown here and on model pages, but it is not counted in any index or ranking until more models have run it.

Score

DLPU phase unwrapping score

Share of tasks passed, 0 to 100 · Higher is better

No data yet.

How to read this chart

What this metric means. The share of tasks the entry passed, 0 to 100.

Data table: DLPU phase unwrapping score
DLPU phase unwrapping score
ModelDeveloperDLPU phase unwrapping score
Claude Sonnet 4.6 (high), Terminus-2AnthropicNo data
DeepSeek V4 Flash (max), Terminus-2DeepSeekNo data
DeepSeek V4 Flash (high), Terminus-2DeepSeekNo data
DeepSeek V4 Pro (max), Terminus-2DeepSeekNo data
DeepSeek V4 Pro (high), Terminus-2DeepSeekNo data
DeepSeek V4.1 Flash (max), Terminus-2DeepSeekNo data
Gemini 3.1 Flash-Lite (low), Terminus-2GoogleNo data
Gemini 3.1 Flash-Lite (minimal), Terminus-2GoogleNo data
Gemini 3.1 Flash-Lite (high), Terminus-2GoogleNo data
Gemini 3.1 Pro (preview) (low), Terminus-2GoogleNo data
Gemini 3.1 Pro (preview) (high), Terminus-2GoogleNo data
Gemini 3.8 Flash (high), Terminus-2GoogleNo data
Gemini 3.8 Flash (low), Terminus-2GoogleNo data
Gemma 4 26B A4B (Reasoning), Terminus-2GoogleNo data
Gemma 4 31B (Reasoning), Terminus-2GoogleNo data
GLM-4.7-Flash (Reasoning), Terminus-2Z.aiNo data
GLM-5.2 (high), Terminus-2Z.aiNo data
GLM-5.2 (Non-reasoning), Terminus-2Z.aiNo data
GLM-5.2 (max), Terminus-2Z.aiNo data
GLM-5.3-Flash (low), Terminus-2Z.aiNo data
GLM-5.3-Flash (max), Terminus-2Z.aiNo data
GLM-5.3 (low), Terminus-2Z.aiNo data
GLM-5.3 (max), Terminus-2Z.aiNo data
gpt-oss-120b (low), Terminus-2OpenAINo data
gpt-oss-120b (high), Terminus-2OpenAINo data
gpt-oss-20b (high), Terminus-2OpenAINo data
Kimi K2.6 (Reasoning), Terminus-2Moonshot AINo data
Kimi K2.7 Code (Reasoning), Terminus-2Moonshot AINo data
Kimi K3 (Reasoning), Terminus-2Moonshot AINo data
Laguna XS 2.1 (Reasoning), Terminus-2PoolsideNo data
Llama 3.3 70B (Non-reasoning), Terminus-2MetaNo data
MiMo V2.6 Flash (Reasoning), Terminus-2XiaomiNo data
MiniMax M3.1 Flash (max), Terminus-2MiniMaxNo data
MiniMax M3 (Reasoning), Terminus-2MiniMaxNo data
Mistral Large 3 (Non-reasoning), Terminus-2Mistral AINo data
Mistral Large 4 (high), Terminus-2Mistral AINo data
Muse Glimmer 30B (high), Terminus-2MetaNo data
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIANo data
Nemotron 3 Ultra (high), Terminus-2NVIDIANo data
Nemotron 3.5 Bolt E4B (Reasoning), Terminus-2NVIDIANo data
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIANo data
Nemotron 3.5 Super 120B (Reasoning), Terminus-2NVIDIANo data
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIANo data
Qwen 3.6 35B-A3B (Reasoning), Terminus-2Alibaba (Qwen)No data
Qwen 3.8 27B (xhigh), Terminus-2Alibaba (Qwen)No data
Step 5 Preview (Reasoning), Terminus-2StepFunNo data
Claude Haiku 4.5 (Reasoning), Claude CodeAnthropicNo data
DeepSeek V4 Flash (high), Claude CodeDeepSeekNo data
DeepSeek V4.1 Flash (max), Claude CodeDeepSeekNo data
GLM-4.7-Flash (Reasoning), Claude CodeZ.aiNo data
GLM-5.2 (max), Claude CodeZ.aiNo data
GLM-5.3-Flash (max), Claude CodeZ.aiNo data
GLM-5.3 (max), Claude CodeZ.aiNo data
DeepSeek V4 Flash (high), CodexDeepSeekNo data
DeepSeek V4 Pro (high), CodexDeepSeekNo data
DeepSeek V4.1 Flash (max), CodexDeepSeekNo data
gpt-oss-120b (high), CodexOpenAINo data
gpt-oss-20b (high), CodexOpenAINo data
Kimi K2.6 (Reasoning), Kimi CodeMoonshot AINo data
Kimi K2.7 Code (Reasoning), Kimi CodeMoonshot AINo data
Kimi K3 (Reasoning), Kimi CodeMoonshot AINo data
MiniMax M3 (Reasoning), MiniMax CodeMiniMaxNo data
MiMo V2.6 Flash (Reasoning), MiMo CodeXiaomiNo data
Qwen 3.6 35B-A3B (Reasoning), Qwen CodeAlibaba (Qwen)No data
Qwen 3.8 27B (Reasoning), Qwen CodeAlibaba (Qwen)No data
Claude Opus 5.5 (medium), Claude CodeAnthropicNo data
Claude Opus 5.5 (high), Claude CodeAnthropicNo data
Claude Sonnet 5.5 (medium), Claude CodeAnthropicNo data
Claude Sonnet 5.5 (high), Claude CodeAnthropicNo data
Claude Haiku 5.5 (medium), Claude CodeAnthropicNo data
Claude Haiku 5.5 (high), Claude CodeAnthropicNo data
Claude Opus 5.5 (xhigh), Claude CodeAnthropicNo data
Claude Sonnet 5.5 (xhigh), Claude CodeAnthropicNo data
Claude Haiku 5.5 (xhigh), Claude CodeAnthropicNo data
Claude Opus 5.5 (low), Claude CodeAnthropicNo data
Claude Sonnet 5.5 (low), Claude CodeAnthropicNo data
Claude Haiku 5.5 (low), Claude CodeAnthropicNo data
Claude Opus 5.5 (max), Claude CodeAnthropicNo data
Claude Sonnet 5.5 (max), Claude CodeAnthropicNo data
Claude Haiku 5.5 (max), Claude CodeAnthropicNo data
Claude Fable 5.1 (medium), Claude CodeAnthropicNo data
Claude Fable 5.1 (high), Claude CodeAnthropicNo data
Claude Opus 5 (high), Claude CodeAnthropicNo data
Claude Sonnet 5 (high), Claude CodeAnthropicNo data
Claude Opus 4.8 (high), Claude CodeAnthropicNo data
Claude Opus 4.7 (high), Claude CodeAnthropicNo data
GPT-6.1 Sol (medium), CodexOpenAINo data
GPT-6.1 Sol (high), CodexOpenAINo data
GPT-6.1 Sol (xhigh), CodexOpenAINo data
GPT-6 Luna (xhigh), CodexOpenAINo data
GPT-6 Sol (medium), CodexOpenAINo data
GPT-6 Sol (high), CodexOpenAINo data
GPT-6 Luna (medium), CodexOpenAINo data
GPT-6 Luna (high), CodexOpenAINo data
GPT-5.6 Sol (medium), CodexOpenAINo data
GPT-5.6 Sol (high), CodexOpenAINo data
GPT-5.6 Terra (high), CodexOpenAINo data
GPT-5.6 Luna (medium), CodexOpenAINo data
GPT-5.6 Luna (high), CodexOpenAINo data
GPT-5.6 Luna (xhigh), CodexOpenAINo data
GPT-6 Luna (low), CodexOpenAINo data
GPT-5.6 Luna (low), CodexOpenAINo data

Background

A radar interferogram records ground height and movement as a phase that wraps around every cycle. Unwrapping puts back the right whole number of cycles at every pixel. These tasks use simulated 256 × 256 interferograms from two public sets, LB-DLPU (L-band, NISAR-like) and InSAR-DLPU (from an SRTM elevation model over China), where the true phase is known.

Licence: both sets CC BY 4.0; LB-DLPU contains modified Copernicus GLO-30 DEM data (© ESA / Copernicus). InSAR-DLPU's real TanDEM-X pairs are not used.

Published by the LB-DLPU authors (Dagbanja, Qian and Lv) and the InSAR-DLPU authors.

Related links

Explore evaluations