PhysDesignBench
GitHub
Reimplementation task · v6-runtime-convergence-diagnostics

TFLN PI-MMI-v0

Polarization-Insensitive TFLN 1x2 MMI Implementation/Optimization

Implement and optimize a polarization-insensitive TFLN 1 × 2 MMI. The trusted evaluator measures worst-output excess loss for TE and TM across 1520–1580 nm.

Runs
11
Scored finals
11 of 11
Best final
S = 0.34 Fable 5.1 · Claude Code
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: Fable 5.1 (Claude Code) 0.335; Opus 5.5 (Claude Code) 0.327; GPT-6 Astra (Codex) 0.326; GPT-6.1 Sol (Codex) 0.321; Sonnet 5.5 (high) (Claude Code) 0.305; Muse Wonder (high) (Muse Code) 0.305; GPT-5.6 Sol (Codex) 0.305; Kimi K3.1 Pro (Claude Code) 0.273; GLM 5.3 Flash (high) (Claude Code) 0.252; Gemini 4 Argon (mini-SWE-agent) 0.224; GLM 5.3 Flash (max) (mini-SWE-agent) 0.120.0.000.250.500.751.000153045607590105120135Elapsed time (min)Score SGLM 5.3 Flash (max) · mini-SWE-agent: final S = 0.120 at 111 minGemini 4 Argon · mini-SWE-agent: final S = 0.224 at 121 minGLM 5.3 Flash (high) · Claude Code: final S = 0.252 at 105 minKimi K3.1 Pro · Claude Code: final S = 0.273 at 121 minGPT-5.6 Sol · Codex: final S = 0.305 at 121 minMuse Wonder (high) · Muse Code: final S = 0.305 at 56 minSonnet 5.5 (high) · Claude Code: final S = 0.305 at 56 minGPT-6.1 Sol · Codex: final S = 0.321 at 120 minGPT-6 Astra · Codex: final S = 0.326 at 120 minOpus 5.5 · Claude Code: final S = 0.327 at 76 minFable 5.1 · Claude Code: final S = 0.335 at 81 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.

  • LN rib

  • TFLN PI-MMI-v0 · Claude Code · Claude Fable 5.1

    S = 0.335 · scale bar 10 µm

    Fit
    Fable 5.1Claude CodeS = 0.34

  • TFLN PI-MMI-v0 · Claude Code · Claude Opus 5.5

    S = 0.327 · scale bar 10 µm

    Fit
    Opus 5.5Claude CodeS = 0.33

  • TFLN PI-MMI-v0 · Codex · GPT-6 Astra

    S = 0.326 · scale bar 10 µm

    Fit
    GPT-6 AstraCodexS = 0.33

  • TFLN PI-MMI-v0 · Codex · GPT-6.1-sol

    S = 0.321 · scale bar 10 µm

    Fit
    GPT-6.1 SolCodexS = 0.32

  • TFLN PI-MMI-v0 · Claude Code · Claude Sonnet 5.5

    S = 0.305 · scale bar 10 µm

    Fit
    Sonnet 5.5 (high)Claude CodeS = 0.31

  • TFLN PI-MMI-v0 · Muse Code · Muse Wonder

    S = 0.305 · scale bar 10 µm

    Fit
    Muse Wonder (high)Muse CodeS = 0.31

  • TFLN PI-MMI-v0 · Codex · GPT-5.6-sol

    S = 0.305 · scale bar 10 µm

    Fit
    GPT-5.6 SolCodexS = 0.30

  • TFLN PI-MMI-v0 · Claude Code · Kimi K3.1 Pro

    S = 0.273 · scale bar 10 µm

    Fit
    Kimi K3.1 ProClaude CodeS = 0.27

  • TFLN PI-MMI-v0 · Claude Code · GLM 5.3 Flash

    S = 0.252 · scale bar 10 µm

    Fit
    GLM 5.3 Flash (high)Claude CodeS = 0.25

  • TFLN PI-MMI-v0 · mini-SWE-agent · Gemini-4-Argon

    S = 0.224 · scale bar 10 µm

    Fit
    Gemini 4 Argonmini-SWE-agentS = 0.22

  • TFLN PI-MMI-v0 · mini-SWE-agent · GLM 5.3 Flash

    S = 0.120 · scale bar 10 µm

    Fit
    GLM 5.3 Flash (max)mini-SWE-agentS = 0.12