PhysDesignBench
GitHub
Reimplementation task · v3-dual-input-canonical-j

Trapezoidal SWG-v0

Trapezoidal SWG 2 x 2 3-dB Coupler Implementation/Optimization

Implement and optimize a trapezoidal subwavelength-grating 2 × 2 coupler. Compare both physical input responses over 1400–1700 nm against the paper’s splitting and excess-loss envelope.

Runs
11
Scored finals
11 of 11
Best final
S = 0.47 GPT-6.1 Sol · 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-6.1 Sol (Codex) 0.466; GPT-6 Astra (Codex) 0.397; Muse Wonder (high) (Muse Code) 0.371; Fable 5.1 (Claude Code) 0.364; Sonnet 5.5 (high) (Claude Code) 0.349; GLM 5.3 Flash (max) (mini-SWE-agent) 0.338; Opus 5.5 (Claude Code) 0.306; Gemini 4 Argon (mini-SWE-agent) 0.282; GPT-5.6 Sol (Codex) 0.264; Kimi K3.1 Pro (Claude Code) 0.254; GLM 5.3 Flash (high) (Claude Code) 0.035.0.000.250.500.751.000153045607590105120135Elapsed time (min)Score SGLM 5.3 Flash (high) · Claude Code: final S = 0.035 at 76 minKimi K3.1 Pro · Claude Code: final S = 0.254 at 121 minGPT-5.6 Sol · Codex: final S = 0.264 at 121 minGemini 4 Argon · mini-SWE-agent: final S = 0.282 at 121 minOpus 5.5 · Claude Code: final S = 0.306 at 65 minGLM 5.3 Flash (max) · mini-SWE-agent: final S = 0.338 at 112 minSonnet 5.5 (high) · Claude Code: final S = 0.349 at 65 minFable 5.1 · Claude Code: final S = 0.364 at 54 minMuse Wonder (high) · Muse Code: final S = 0.371 at 45 minGPT-6 Astra · Codex: final S = 0.397 at 120 minGPT-6.1 Sol · Codex: final S = 0.466 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

  • Trapezoidal SWG-v0 · Codex · GPT-6.1-sol

    S = 0.466 · scale bar 5 µm

    Fit
    GPT-6.1 SolCodexS = 0.47

  • Trapezoidal SWG-v0 · Codex · GPT-6 Astra

    S = 0.397 · scale bar 5 µm

    Fit
    GPT-6 AstraCodexS = 0.40

  • Trapezoidal SWG-v0 · Muse Code · Muse Wonder

    S = 0.371 · scale bar 5 µm

    Fit
    Muse Wonder (high)Muse CodeS = 0.37

  • Trapezoidal SWG-v0 · Claude Code · Claude Fable 5.1

    S = 0.364 · scale bar 5 µm

    Fit
    Fable 5.1Claude CodeS = 0.36

  • Trapezoidal SWG-v0 · Claude Code · Claude Sonnet 5.5

    S = 0.349 · scale bar 5 µm

    Fit
    Sonnet 5.5 (high)Claude CodeS = 0.35

  • Trapezoidal SWG-v0 · mini-SWE-agent · GLM 5.3 Flash

    S = 0.338 · scale bar 5 µm

    Fit
    GLM 5.3 Flash (max)mini-SWE-agentS = 0.34

  • Trapezoidal SWG-v0 · Claude Code · Claude Opus 5.5

    S = 0.306 · scale bar 5 µm

    Fit
    Opus 5.5Claude CodeS = 0.31

  • Trapezoidal SWG-v0 · mini-SWE-agent · Gemini-4-Argon

    S = 0.282 · scale bar 5 µm

    Fit
    Gemini 4 Argonmini-SWE-agentS = 0.28

  • Trapezoidal SWG-v0 · Codex · GPT-5.6-sol

    S = 0.264 · scale bar 5 µm

    Fit
    GPT-5.6 SolCodexS = 0.26

  • Trapezoidal SWG-v0 · Claude Code · Kimi K3.1 Pro

    S = 0.254 · scale bar 5 µm

    Fit
    Kimi K3.1 ProClaude CodeS = 0.25

  • Trapezoidal SWG-v0 · Claude Code · GLM 5.3 Flash

    S = 0.035 · scale bar 5 µm

    Fit
    GLM 5.3 Flash (high)Claude CodeS = 0.04