PhysDesignBench
GitHub
Reimplementation task · v4-y-mirror-input-coverage

Anisotropic SWG-v0

Anisotropic SWG 2 x 2 Beamsplitter Implementation/Optimization

Implement and optimize an anisotropic subwavelength-grating 2 × 2 MMI beamsplitter over 1250–1750 nm. One input is simulated; strict y-mirror symmetry establishes the second input response.

Runs
11
Scored finals
11 of 11
Best final
S = 0.71 Gemini 4 Argon · mini-SWE-agent
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: Gemini 4 Argon (mini-SWE-agent) 0.705; GPT-6 Astra (Codex) 0.677; Muse Wonder (high) (Muse Code) 0.669; GPT-6.1 Sol (Codex) 0.640; GPT-5.6 Sol (Codex) 0.639; Opus 5.5 (Claude Code) 0.633; Sonnet 5.5 (high) (Claude Code) 0.599; Fable 5.1 (Claude Code) 0.389; Kimi K3.1 Pro (Claude Code) 0.366; GLM 5.3 Flash (high) (Claude Code) 0.317; GLM 5.3 Flash (max) (mini-SWE-agent) 0.312.0.000.250.500.751.000153045607590105120135Elapsed time (min)Score SGLM 5.3 Flash (max) · mini-SWE-agent: final S = 0.312 at 82 minGLM 5.3 Flash (high) · Claude Code: final S = 0.317 at 111 minKimi K3.1 Pro · Claude Code: final S = 0.366 at 121 minFable 5.1 · Claude Code: final S = 0.389 at 118 minSonnet 5.5 (high) · Claude Code: final S = 0.599 at 71 minOpus 5.5 · Claude Code: final S = 0.633 at 75 minGPT-5.6 Sol · Codex: final S = 0.639 at 120 minGPT-6.1 Sol · Codex: final S = 0.640 at 120 minMuse Wonder (high) · Muse Code: final S = 0.669 at 114 minGPT-6 Astra · Codex: final S = 0.677 at 120 minGemini 4 Argon · mini-SWE-agent: final S = 0.705 at 122 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.

  • Silicon
  • Bottom air-void footprint

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

    S = 0.705 · scale bar 5 µm

    Fit
    Gemini 4 Argonmini-SWE-agentS = 0.71

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

    S = 0.677 · scale bar 5 µm

    Fit
    GPT-6 AstraCodexS = 0.68

  • Anisotropic SWG-v0 · Muse Code · Muse Wonder

    S = 0.669 · scale bar 5 µm

    Fit
    Muse Wonder (high)Muse CodeS = 0.67

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

    S = 0.640 · scale bar 5 µm

    Fit
    GPT-6.1 SolCodexS = 0.64

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

    S = 0.639 · scale bar 5 µm

    Fit
    GPT-5.6 SolCodexS = 0.64

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

    S = 0.633 · scale bar 5 µm

    Fit
    Opus 5.5Claude CodeS = 0.63

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

    S = 0.599 · scale bar 5 µm

    Fit
    Sonnet 5.5 (high)Claude CodeS = 0.60

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

    S = 0.389 · scale bar 5 µm

    Fit
    Fable 5.1Claude CodeS = 0.39

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

    S = 0.366 · scale bar 5 µm

    Fit
    Kimi K3.1 ProClaude CodeS = 0.37

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

    S = 0.317 · scale bar 5 µm

    Fit
    GLM 5.3 Flash (high)Claude CodeS = 0.32

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

    S = 0.312 · scale bar 5 µm

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