PhysDesignBench
GitHub
Reimplementation task · v1-binary-paper-envelope-j

HMDM six-channel WDM

HMDM Six-Channel WDM Binary Optimization

Optimize a six-channel WDM nanohole mask with independently selectable sites across the paper-derived design region. The trusted evaluator measures worst-channel insertion loss and crosstalk against the paper envelope.

Runs
11
Scored finals
11 of 11
Best final
S = 0.31 GPT-6 Astra · Codex
Score over timeBest-so-far score S for each run · small dots are probes, large dots final submissions
Score S from 0 to 1 against elapsed minutes. Final scores: GPT-6 Astra (Codex) 0.310; GPT-6.1 Sol (Codex) 0.309; GLM 5.3 Flash (max) (mini-SWE-agent) 0.306; Fable 5.1 (Claude Code) 0.252; Muse Wonder (high) (Muse Code) 0.248; Gemini 4 Argon (mini-SWE-agent) 0.240; GLM 5.3 Flash (high) (Claude Code) 0.214; Kimi K3.1 Pro (Claude Code) 0.153; GPT-5.6 Sol (Codex) 0.063; Opus 5.5 (Claude Code) 0.050; Sonnet 5.5 (high) (Claude Code) 0.028.0.000.250.500.751.000153045607590105120135Elapsed time (min)Score SSonnet 5.5 (high) · Claude Code: final S = 0.028 at 116 minOpus 5.5 · Claude Code: final S = 0.050 at 108 minGPT-5.6 Sol · Codex: final S = 0.063 at 120 minKimi K3.1 Pro · Claude Code: final S = 0.153 at 122 minGLM 5.3 Flash (high) · Claude Code: final S = 0.214 at 121 minGemini 4 Argon · mini-SWE-agent: final S = 0.240 at 121 minMuse Wonder (high) · Muse Code: final S = 0.248 at 118 minFable 5.1 · Claude Code: final S = 0.252 at 97 minGLM 5.3 Flash (max) · mini-SWE-agent: final S = 0.306 at 99 minGPT-6.1 Sol · Codex: final S = 0.309 at 120 minGPT-6 Astra · Codex: final S = 0.310 at 120 min

Showing 11 of 11 runs. Hover a line or a model to isolate it; click a model to hide or show it.

Final layouts

Every run’s final GDS, at one scale.

The design each run ended with, drawn from its evaluated GDS, with the task’s ports, modes and band as the evaluator defines them, the run’s own length, and the published result that S = 0.5 stands for. Every layout uses the same physical scale, so sizes compare directly. Hatched layouts failed a validity gate and score zero. Select a layout to zoom and toggle its layers.

  • Optical mask

  • HMDM six-channel WDM · Codex · GPT-6 Astra

    S = 0.310 · scale bar 2 µm

    Fit
    GPT-6 AstraCodexS = 0.31

  • HMDM six-channel WDM · Codex · GPT-6.1-sol

    S = 0.309 · scale bar 2 µm

    Fit
    GPT-6.1 SolCodexS = 0.31

  • HMDM six-channel WDM · mini-SWE-agent · GLM 5.3 Flash

    S = 0.306 · scale bar 2 µm

    Fit
    GLM 5.3 Flash (max)mini-SWE-agentS = 0.31

  • HMDM six-channel WDM · Claude Code · Claude Fable 5.1

    S = 0.252 · scale bar 2 µm

    Fit
    Fable 5.1Claude CodeS = 0.25

  • HMDM six-channel WDM · Muse Code · Muse Wonder

    S = 0.248 · scale bar 2 µm

    Fit
    Muse Wonder (high)Muse CodeS = 0.25

  • HMDM six-channel WDM · mini-SWE-agent · Gemini-4-Argon

    S = 0.240 · scale bar 2 µm

    Fit
    Gemini 4 Argonmini-SWE-agentS = 0.24

  • HMDM six-channel WDM · Claude Code · GLM 5.3 Flash

    S = 0.214 · scale bar 2 µm

    Fit
    GLM 5.3 Flash (high)Claude CodeS = 0.21

  • HMDM six-channel WDM · Claude Code · Kimi K3.1 Pro

    S = 0.153 · scale bar 2 µm

    Fit
    Kimi K3.1 ProClaude CodeS = 0.15

  • HMDM six-channel WDM · Codex · GPT-5.6-sol

    S = 0.063 · scale bar 2 µm

    Fit
    GPT-5.6 SolCodexS = 0.06

  • HMDM six-channel WDM · Claude Code · Claude Opus 5.5

    S = 0.050 · scale bar 2 µm

    Fit
    Opus 5.5Claude CodeS = 0.05

  • HMDM six-channel WDM · Claude Code · Claude Sonnet 5.5

    S = 0.028 · scale bar 2 µm

    Fit
    Sonnet 5.5 (high)Claude CodeS = 0.03