Physical-Engineering's Last Exam (PLE)
We show this table for reference; we do not rank on it.
Long, end-to-end engineering projects: an agent works through a full project folder, finds the governing codes and physical limits, and produces deliverables a licensed engineer would sign off on.
Evaluation results published by Google.
Score (hidden test set) on PLE — November 12, 2026 capture
We mirror the published score (hidden test set) view for PLE. Claude Opus 5.5 leads the public snapshot at 48.6%, followed by Claude Opus 5.5 (48.3%) and Claude Opus 5.5 (47.7%). We do not use these results to rank models overall.
1Closed
Claude Opus 5.5
Anthropic
openhands
openhands harness · ±2.3 across runs · $14.87 per task · 66 min per task · Structural 57.6, Civil 64, Electrical 47.9, Mechanical 40.8
48.6%
Overall 86.62Context 1M
2Closed
Claude Opus 5.5
Anthropic
claude-code
claude-code harness (native) · ±1.4 across runs · $17.12 per task · 63 min per task · Structural 56.3, Civil 70.3, Electrical 43.3, Mechanical 41.6
48.3%
Overall 86.62Context 1M
3Closed
Claude Opus 5.5
Anthropic
agy
agy harness · ±0.8 across runs · about $42.32 per task · 45 min per task · Structural 49.6, Civil 62.1, Electrical 46.8, Mechanical 42.4
47.7%
Overall 86.62Context 1M
122 model and harness pairsAgenticCurrentDisplay onlyUpdated November 12, 2026 capture
Score (hidden test set) table (122 model and harness pairs)
122 of 122 rows
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropicopenhandsopenhands harness · ±2.3 across runs · $14.87 per task · 66 min per task · Structural 57.6, Civil 64, Electrical 47.9, Mechanical 40.8 | 48.6% | Not reported | Closed |
| 2 | Claude Opus 5.5Anthropicclaude-codeclaude-code harness (native) · ±1.4 across runs · $17.12 per task · 63 min per task · Structural 56.3, Civil 70.3, Electrical 43.3, Mechanical 41.6 | 48.3% | Not reported | Closed |
| 3 | Claude Opus 5.5Anthropicagyagy harness · ±0.8 across runs · about $42.32 per task · 45 min per task · Structural 49.6, Civil 62.1, Electrical 46.8, Mechanical 42.4 | 47.7% | Not reported | Closed |
| 4 | Claude Opus 5.5Anthropicpipi harness · ±1.4 across runs · $13.35 per task · 68 min per task · Structural 54.6, Civil 62.4, Electrical 45.3, Mechanical 40.9 | 47.2% | Not reported | Closed |
| 5 | Claude Opus 5.5Anthropichermeshermes harness · ±0.5 across runs · $16.03 per task · 66 min per task · Structural 47.2, Civil 62, Electrical 49, Mechanical 38.5 | 46.6% | Not reported | Closed |
| 6 | Claude Opus 5.5Anthropicopencodeopencode harness · ±2 across runs · $12.87 per task · 49 min per task · Structural 53.9, Civil 62.3, Electrical 43.8, Mechanical 40.5 | 46.5% | Not reported | Closed |
| 7 | Claude Opus 5.5Anthropiccodexcodex harness · ±1.9 across runs · $37.62 per task · 130 min per task · Structural 51.6, Civil 63.5, Electrical 48.5, Mechanical 36.7 | 46.4% | Not reported | Closed |
| 8 | GPT-6 AstraOpenAIcodexcodex harness (native) · ±0.6 across runs · $9.53 per task · 26 min per task · Structural 48.4, Civil 57.6, Electrical 40.4, Mechanical 44.8 | 45.8% | Not reported | Closed |
| 9 | Claude Sonnet 5.5Anthropichermeshermes harness · ±3.8 across runs · $14.71 per task · 87 min per task · Structural 51.8, Civil 55.7, Electrical 40.8, Mechanical 42.5 | 45.1% | Not reported | Closed |
| 10 | Claude Sonnet 5.5Anthropicagyagy harness · ±0.4 across runs · about $33.89 per task · 53 min per task · Structural 50.1, Civil 56.2, Electrical 39.5, Mechanical 43.2 | 44.9% | Not reported | Closed |
| 11 | Claude Sonnet 5.5Anthropicpipi harness · ±2.2 across runs · $11.77 per task · 96 min per task · Structural 48.9, Civil 57.9, Electrical 39.9, Mechanical 40.9 | 44.2% | Not reported | Closed |
| 12 | Claude Sonnet 5.5Anthropicclaude-codeclaude-code harness (native) · ±1.4 across runs · $18.68 per task · 99 min per task · Structural 47.4, Civil 52, Electrical 40.6, Mechanical 41.5 | 43.5% | Not reported | Closed |
| 13 | Claude Sonnet 5.5Anthropiccodexcodex harness · ±2.9 across runs · $42.56 per task · 243 min per task · Structural 46.7, Civil 51.4, Electrical 42.1, Mechanical 39.7 | 43.1% | Not reported | Closed |
| 14 | Claude Opus 5Anthropicopenhandsopenhands harness · ±0.9 across runs · $17.13 per task · 66 min per task · Structural 52, Civil 52.5, Electrical 37.8, Mechanical 40 | 42.6% | Not reported | Closed |
| 15 | Claude Sonnet 5.5Anthropicopencodeopencode harness · ±0.7 across runs · $10.53 per task · 63 min per task · Structural 52.2, Civil 52.1, Electrical 39, Mechanical 38.9 | 42.5% | Not reported | Closed |
| 16 | Claude Opus 5Anthropicpipi harness · ±1 across runs · $15.26 per task · 58 min per task · Structural 51.3, Civil 43.1, Electrical 40.6, Mechanical 39.8 | 41.8% | Not reported | Closed |
| 17 | Claude Opus 5Anthropicagyagy harness · ±2.1 across runs · about $29.59 per task · 42 min per task · Structural 52.1, Civil 45.5, Electrical 37.4, Mechanical 40.5 | 41.5% | Not reported | Closed |
| 18 | Gemini 4 ArgonGoogleclaude-codeclaude-code harness · ±1.4 across runs · $8.51 per task · 67 min per task · Structural 44.9, Civil 51.5, Electrical 41, Mechanical 36.1 | 41.1% | Not reported | Closed |
| 19 | Claude Sonnet 5.5Anthropicopenhandsopenhands harness · ±1.4 across runs · $15.72 per task · 112 min per task · Structural 51.7, Civil 47, Electrical 36, Mechanical 39.1 | 40.7% | Not reported | Closed |
| 20 | Gemini 4 ArgonGoogleagyagy harness (native) · ±1.3 across runs · $7.7 per task · 73 min per task · Structural 45.8, Civil 53.8, Electrical 39.7, Mechanical 34.8 | 40.6% | Not reported | Closed |
| 21 | Claude Opus 5Anthropichermeshermes harness · ±1.7 across runs · $13.89 per task · 52 min per task · Structural 49.7, Civil 44.3, Electrical 37.9, Mechanical 38.4 | 40.4% | Not reported | Closed |
| 22 | Claude Opus 5Anthropicopencodeopencode harness · ±1.7 across runs · $12.51 per task · 38 min per task · Structural 48.7, Civil 45.6, Electrical 35.8, Mechanical 39.6 | 40.4% | Not reported | Closed |
| 23 | Claude Opus 5Anthropicclaude-codeclaude-code harness (native) · ±1.6 across runs · $16.52 per task · 48 min per task · Structural 47.3, Civil 46.1, Electrical 36.7, Mechanical 38.3 | 40.0% | Not reported | Closed |
| 24 | Claude Opus 5Anthropiccodexcodex harness · ±0.6 across runs · $20.77 per task · 69 min per task · Structural 52.3, Civil 39.4, Electrical 36.5, Mechanical 37 | 38.9% | Not reported | Closed |
| 25 | Gemini 4 ArgonGooglecodexcodex harness · ±1.7 across runs · $7.13 per task · 83 min per task · Structural 39.9, Civil 54.5, Electrical 34.6, Mechanical 35.9 | 38.9% | Not reported | Closed |
| 26 | Gemini 4 ArgonGooglehermeshermes harness · ±1.2 across runs · $7.92 per task · 61 min per task · Structural 38.2, Civil 48.1, Electrical 40.7, Mechanical 33.9 | 38.9% | Not reported | Closed |
| 27 | Gemini 4 ArgonGoogleopencodeopencode harness · ±0.6 across runs · $7.04 per task · 55 min per task · Structural 43.8, Civil 47, Electrical 34.1, Mechanical 34.3 | 37.3% | Not reported | Closed |
| 28 | Gemini 4 ArgonGooglepipi harness · ±1.5 across runs · $7 per task · 78 min per task · Structural 42, Civil 50.3, Electrical 32.3, Mechanical 33.1 | 36.6% | Not reported | Closed |
| 29 | Claude Fable 5Anthropicclaude-codeclaude-code harness (native) · ±2 across runs · $26.46 per task · 45 min per task · Structural 40.6, Civil 39.1, Electrical 31.7, Mechanical 33.2 | 34.5% | Not reported | Closed |
| 30 | Claude Fable 5Anthropiccodexcodex harness · ±1.5 across runs · $28.77 per task · 60 min per task · Structural 48, Civil 35.1, Electrical 31.2, Mechanical 33.3 | 34.5% | Not reported | Closed |
| 31 | Claude Fable 5Anthropicpipi harness · ±1.6 across runs · $21.17 per task · 47 min per task · Structural 46.9, Civil 31.5, Electrical 35, Mechanical 31.4 | 34.2% | Not reported | Closed |
| 32 | Claude Fable 5Anthropicopencodeopencode harness · ±1.5 across runs · $22.95 per task · 38 min per task · Structural 39.6, Civil 32.6, Electrical 33.2, Mechanical 33.9 | 34.1% | Not reported | Closed |
| 33 | Claude Fable 5Anthropichermeshermes harness · ±0.9 across runs · $24.3 per task · 55 min per task · Structural 44.6, Civil 31.2, Electrical 32, Mechanical 32.3 | 33.3% | Not reported | Closed |
| 34 | Claude Fable 5Anthropicopenhandsopenhands harness · ±0.4 across runs · $25.83 per task · 44 min per task · Structural 48.7, Civil 32.6, Electrical 31.4, Mechanical 30.9 | 33.2% | Not reported | Closed |
| 35 | Claude Fable 5Anthropicagyagy harness · ±1 across runs · about $54.19 per task · 37 min per task · Structural 37.6, Civil 31.9, Electrical 32.3, Mechanical 31.5 | 32.5% | Not reported | Closed |
| 36 | Grok 4.7xAIpipi harness · ±0.5 across runs · $8.01 per task · 40 min per task · Structural 32.5, Civil 34.6, Electrical 30.9, Mechanical 27.3 | 30.2% | Not reported | Closed |
| 37 | Grok 4.7xAIgrok-buildgrok-build harness (native) · ±0.7 across runs · $10.18 per task · 52 min per task · Structural 28.3, Civil 37.1, Electrical 31.2, Mechanical 27 | 30.1% | Not reported | Closed |
| 38 | Gemini 4 ArgonGoogleopenhandsopenhands harness · ±3.3 across runs · $7.54 per task · 83 min per task · Structural 32.2, Civil 29.1, Electrical 31.8, Mechanical 26.1 | 29.0% | Not reported | Closed |
| 39 | Grok 4.7xAIopencodeopencode harness · ±1.7 across runs · $10.25 per task · 34 min per task · Structural 28.9, Civil 35.9, Electrical 26.9, Mechanical 26.5 | 28.4% | Not reported | Closed |
| 40 | Gemini 3.8 FlashGoogleopenhandsopenhands harness · ±1 across runs · $3.21 per task · 33 min per task · Structural 33.9, Civil 30.9, Electrical 26.2, Mechanical 27.1 | 28.2% | Not reported | Closed |
| 41 | Gemini 3.8 FlashGooglecodexcodex harness · ±0.1 across runs · $3.61 per task · 45 min per task · Structural 28.4, Civil 30.1, Electrical 25.9, Mechanical 28.2 | 27.8% | Not reported | Closed |
| 42 | Grok 4.7xAIagyagy harness · ±0.7 across runs · about $16.81 per task (source footnote) · 33 min per task · Structural 26.4, Civil 32.1, Electrical 29.6, Mechanical 25.1 | 27.8% | Not reported | Closed |
| 43 | Gemini 3.8 FlashGoogleagyagy harness (native) · ±0.7 across runs · $2.6 per task · 40 min per task · Structural 27.1, Civil 31.8, Electrical 26.2, Mechanical 26.9 | 27.5% | Not reported | Closed |
| 44 | Grok 4.7xAIopenhandsopenhands harness · ±1.9 across runs · $8.89 per task · 34 min per task · Structural 27.3, Civil 28.6, Electrical 32.8, Mechanical 23 | 27.5% | Not reported | Closed |
| 45 | Gemini 3.8 FlashGoogleclaude-codeclaude-code harness · ±0.3 across runs · $2.61 per task · 34 min per task · Structural 32.2, Civil 28.6, Electrical 23.2, Mechanical 28.5 | 27.2% | Not reported | Closed |
| 46 | Gemini 3.8 FlashGooglepipi harness · ±2.2 across runs · $2.68 per task · 42 min per task · Structural 25.6, Civil 28.9, Electrical 27.1, Mechanical 26.1 | 26.8% | Not reported | Closed |
| 47 | Grok 4.7xAIclaude-codeclaude-code harness · ±1.2 across runs · $12.75 per task · 48 min per task · Structural 26.5, Civil 26, Electrical 27.1, Mechanical 26.3 | 26.6% | Not reported | Closed |
| 48 | Gemini 3.8 FlashGoogleopencodeopencode harness · ±0.5 across runs · $2.45 per task · 25 min per task · Structural 25.3, Civil 25.5, Electrical 25, Mechanical 26.9 | 25.9% | Not reported | Closed |
| 49 | Grok 4.7xAIcodexcodex harness · ±2 across runs · $7.02 per task · 39 min per task · Structural 26.4, Civil 31.4, Electrical 28.1, Mechanical 21.9 | 25.9% | Not reported | Closed |
| 50 | Grok 4.7xAIhermeshermes harness · ±2.1 across runs · $10.09 per task · 47 min per task · Structural 24.5, Civil 29.7, Electrical 25.6, Mechanical 24.3 | 25.6% | Not reported | Closed |
| 51 | GLM-5.2Z.AIcodexcodex harness · ±1.5 across runs · $12.57 per task · 121 min per task · Structural 26.8, Civil 23.8, Electrical 24.3, Mechanical 25.6 | 25.0% | Not reported | Open |
| 52 | GLM-5.2Z.AIpipi harness · ±0.4 across runs · $13.66 per task · 109 min per task · Structural 26.3, Civil 26.3, Electrical 22.3, Mechanical 26.3 | 25.0% | Not reported | Open |
| 53 | GLM-5.2Z.AIhermeshermes harness · ±1.2 across runs · $10.83 per task · 70 min per task · Structural 24.9, Civil 23.4, Electrical 22.6, Mechanical 25.1 | 24.0% | Not reported | Open |
| 54 | GLM-5.2Z.AIopencodeopencode harness · ±1.1 across runs · Cost not reported · 94 min per task · Structural 27.6, Civil 20.9, Electrical 22.9, Mechanical 25.2 | 24.0% | Not reported | Open |
| 55 | GLM-5.2Z.AIagyagy harness · ±0.8 across runs · about $12.71 per task · 86 min per task · Structural 22.3, Civil 23.4, Electrical 23.3, Mechanical 24.7 | 23.8% | Not reported | Open |
| 56 | GLM-5.2Z.AIopenhandsopenhands harness · ±0.5 across runs · $15.5 per task · 104 min per task · Structural 21.1, Civil 23.3, Electrical 22.7, Mechanical 24.4 | 23.4% | Not reported | Open |
| 57 | DeepSeek-V4.1-FlashDeepSeekopencodeopencode harness · ±2.3 across runs · $3.46 per task · 172 min per task · Structural 32.1, Civil 21.7, Electrical 20.1, Mechanical 22.5 | 22.5% | Not reported | Unknown |
| 58 | DeepSeek-V4.1-FlashDeepSeekpipi harness · ±3.4 across runs · $4.54 per task · 169 min per task · Structural 21.4, Civil 26.8, Electrical 21.1, Mechanical 22.3 | 22.5% | Not reported | Unknown |
| 59 | GLM-5.2Z.AIclaude-codeclaude-code harness · ±0.5 across runs · $14.78 per task · 94 min per task · Structural 23.1, Civil 20.6, Electrical 22.2, Mechanical 21.7 | 21.8% | Not reported | Open |
| 60 | Gemini 3.8 FlashGooglehermeshermes harness · ±2 across runs · $2.39 per task · 24 min per task · Structural 11.7, Civil 18.4, Electrical 18.6, Mechanical 17.7 | 17.4% | Not reported | Closed |
| 61 | Gemini 3.1 ProGooglehermeshermes harness · ±1.6 across runs · $2.09 per task · 12 min per task · Structural 11.7, Civil 14.8, Electrical 16.2, Mechanical 14.6 | 14.8% | Not reported | Closed |
| 62 | Gemini 3.1 ProGoogleopenhandsopenhands harness · ±1.6 across runs · $4.67 per task · 16 min per task · Structural 9.8, Civil 14.4, Electrical 16, Mechanical 15.2 | 14.7% | Not reported | Closed |
| 63 | Gemini 3.1 ProGooglecodexcodex harness · ±0.2 across runs · $2.8 per task · 16 min per task · Structural 11.4, Civil 15.2, Electrical 16, Mechanical 13.9 | 14.5% | Not reported | Closed |
| 64 | DeepSeek-V4.1-FlashDeepSeekdshdsh harness (native) · ±2.3 across runs · $6.3 per task · 240 min per task · Structural 1.7, Civil 9, Electrical 14.1, Mechanical 20.1 | 14.4% | Not reported | Unknown |
| 65 | Gemini 3.1 ProGoogleclaude-codeclaude-code harness · ±0.6 across runs · $3.65 per task · 17 min per task · Structural 12.9, Civil 13.3, Electrical 15.9, Mechanical 14 | 14.4% | Not reported | Closed |
| 66 | Gemini 3.1 ProGoogleopencodeopencode harness · ±0.5 across runs · $2.67 per task · 15 min per task · Structural 11.4, Civil 13.5, Electrical 15.3, Mechanical 14.1 | 14.1% | Not reported | Closed |
| 67 | Gemini 3.1 ProGooglepipi harness · ±0.8 across runs · $2.76 per task · 24 min per task · Structural 11.9, Civil 12.1, Electrical 15.1, Mechanical 14.6 | 14.1% | Not reported | Closed |
| 68 | Gemini 3.1 ProGoogleagyagy harness (native) · ±1.4 across runs · $2.47 per task · 15 min per task · Structural 7.8, Civil 13.2, Electrical 15.8, Mechanical 14.1 | 13.8% | Not reported | Closed |
| 69 | DeepSeek-V4.1-FlashDeepSeekclaude-codeclaude-code harness · ±1.2 across runs · $3.74 per task · 241 min per task · Structural 3.3, Civil 11.6, Electrical 14.5, Mechanical 11.5 | 11.7% | Not reported | Unknown |
| 70 | DeepSeek-V4.1-FlashDeepSeekagyagy harness · ±1.3 across runs · about $3.5 per task · 229 min per task · Structural 1.7, Civil 6.2, Electrical 10, Mechanical 11.7 | 9.2% | Not reported | Unknown |
| 71 | DeepSeek-V4.1-FlashDeepSeekopenhandsopenhands harness · ±1.5 across runs · $1.65 per task · 194 min per task · Structural 0, Civil 5.8, Electrical 4.2, Mechanical 9.7 | 6.3% | Not reported | Unknown |
| 72 | Kimi K2 ThinkingMoonshot AIopencodeopencode harness · ±0.2 across runs · $0.28 per task · 8 min per task · Structural 4.5, Civil 3.2, Electrical 9.1, Mechanical 5.4 | 6.2% | Not reported | Unknown |
| 73 | Qwen3 Coder 480BAlibabaclaude-codeclaude-code harness · ±0.1 across runs · $0.13 per task · 10 min per task · Structural 2.2, Civil 3.6, Electrical 9, Mechanical 6.2 | 6.2% | Not reported | Unknown |
| 74 | Qwen3 Coder 480BAlibabahermeshermes harness · ±0.4 across runs · $0.12 per task · 8 min per task · Structural 2.7, Civil 2.2, Electrical 8.5, Mechanical 6.9 | 6.2% | Not reported | Unknown |
| 75 | Qwen3 Coder 480BAlibabaopenhandsopenhands harness · ±0.7 across runs · $0.21 per task · 28 min per task · Structural 3.1, Civil 2.1, Electrical 8.5, Mechanical 6.4 | 6.0% | Not reported | Unknown |
| 76 | MiniMax M2MiniMaxclaude-codeclaude-code harness · ±0.6 across runs · $0.1 per task · 12 min per task · Structural 2.7, Civil 2.9, Electrical 8.8, Mechanical 5.1 | 5.7% | Not reported | Unknown |
| 77 | MiniMax M2MiniMaxhermeshermes harness · ±0.5 across runs · $0.097 per task · 10 min per task · Structural 2.3, Civil 4, Electrical 6.8, Mechanical 6.3 | 5.7% | Not reported | Unknown |
| 78 | DeepSeek-V4.1-FlashDeepSeekhermeshermes harness · ±2.6 across runs · $2.98 per task · 255 min per task · Structural 0, Civil 3.2, Electrical 5.1, Mechanical 7.7 | 5.3% | Not reported | Unknown |
| 79 | MiniMax M2MiniMaxopenhandsopenhands harness · ±0.3 across runs · $0.13 per task · 14 min per task · Structural 0.2, Civil 2.3, Electrical 9.6, Mechanical 4.4 | 5.3% | Not reported | Unknown |
| 80 | Kimi K2 ThinkingMoonshot AIhermeshermes harness · ±0.6 across runs · $0.28 per task · 8 min per task · Structural 1, Civil 3.4, Electrical 7.5, Mechanical 4.6 | 5.0% | Not reported | Unknown |
| 81 | Qwen3 Coder 480BAlibabacodexcodex harness · ±0.6 across runs · $0.11 per task · 6 min per task · Structural 2, Civil 3.1, Electrical 6.7, Mechanical 4.4 | 4.7% | Not reported | Unknown |
| 82 | GPT-OSS 120BOpenAIopenhandsopenhands harness · ±0.9 across runs · $0.14 per task · 52 min per task · Structural 1.7, Civil 2, Electrical 5.4, Mechanical 5.7 | 4.6% | Not reported | Open |
| 83 | Kimi K2 ThinkingMoonshot AIclaude-codeclaude-code harness · ±1.4 across runs · $0.25 per task · 7 min per task · Structural 0.2, Civil 2.8, Electrical 7.6, Mechanical 4 | 4.6% | Not reported | Unknown |
| 84 | MiniMax M2MiniMaxcodexcodex harness · ±0.2 across runs · $0.093 per task · 14 min per task · Structural 2.8, Civil 2.2, Electrical 6.9, Mechanical 4.3 | 4.6% | Not reported | Unknown |
| 85 | MiniMax M2MiniMaxpipi harness · ±0.2 across runs · $0.094 per task · 12 min per task · Structural 1.2, Civil 3, Electrical 7.8, Mechanical 3.5 | 4.6% | Not reported | Unknown |
| 86 | Gemma 4 26B A4BGoogleopenhandsopenhands harness · ±0.5 across runs · $0.61 per task · 109 min per task · Structural 0, Civil 2.6, Electrical 4.6, Mechanical 6.3 | 4.5% | Not reported | Open |
| 87 | Qwen3 Coder 480BAlibabaopencodeopencode harness · ±0.2 across runs · $0.25 per task · 8 min per task · Structural 0.1, Civil 1.9, Electrical 7.2, Mechanical 4.6 | 4.5% | Not reported | Unknown |
| 88 | Kimi K2 ThinkingMoonshot AIopenhandsopenhands harness · ±0.7 across runs · $0.25 per task · 12 min per task · Structural 0.1, Civil 2.7, Electrical 5.8, Mechanical 5.1 | 4.4% | Not reported | Unknown |
| 89 | Qwen3 Coder 480BAlibabaagyagy harness · ±0.6 across runs · about $0.49 per task · 6 min per task · Structural 0.7, Civil 3.6, Electrical 5.8, Mechanical 4.6 | 4.4% | Not reported | Unknown |
| 90 | Qwen3-Next 80BAlibabaopenhandsopenhands harness · ±0.1 across runs · $0.77 per task (source footnote) · 92 min per task · Structural 1.1, Civil 2, Electrical 6.5, Mechanical 4.4 | 4.3% | Not reported | Unknown |
| 91 | Gemma 4 26B A4BGoogleagyagy harness (native) · ±0.4 across runs · about $0.21 per task · 95 min per task · Structural 0.7, Civil 2.4, Electrical 4.2, Mechanical 5.8 | 4.2% | Not reported | Open |
| 92 | GPT-OSS 120BOpenAIagyagy harness · ±0.3 across runs · Cost not reported · 36 min per task · Structural 1.1, Civil 1.1, Electrical 6.1, Mechanical 4.7 | 4.2% | Not reported | Open |
| 93 | GPT-OSS 120BOpenAIclaude-codeclaude-code harness · ±0.2 across runs · $0.11 per task · 46 min per task · Structural 0.5, Civil 2.1, Electrical 5.4, Mechanical 4.9 | 4.2% | Not reported | Open |
| 94 | MiniMax M2MiniMaxopencodeopencode harness · ±0.4 across runs · $0.079 per task · 14 min per task · Structural 0, Civil 3.1, Electrical 7.6, Mechanical 3 | 4.2% | Not reported | Unknown |
| 95 | Qwen3 Coder 480BAlibabapipi harness · ±0.5 across runs · $0.15 per task · 10 min per task · Structural 0.9, Civil 2.3, Electrical 7, Mechanical 3.7 | 4.2% | Not reported | Unknown |
| 96 | DeepSeek-V4.1-FlashDeepSeekcodexcodex harness · ±1.6 across runs · $2.08 per task · 295 min per task · Structural 2, Civil 3.7, Electrical 3, Mechanical 5.5 | 4.1% | Not reported | Unknown |
| 97 | GPT-OSS 120BOpenAIpipi harness · ±0.6 across runs · $0.12 per task · 26 min per task · Structural 1.5, Civil 3.3, Electrical 5.6, Mechanical 3.5 | 4.0% | Not reported | Open |
| 98 | Kimi K2 ThinkingMoonshot AIpipi harness · ±0.2 across runs · $0.2 per task · 8 min per task · Structural 0, Civil 3, Electrical 5.2, Mechanical 4.5 | 4.0% | Not reported | Unknown |
| 99 | Gemma 4 26B A4BGoogleclaude-codeclaude-code harness · ±0.2 across runs · $0.53 per task · 88 min per task · Structural 0.1, Civil 2.5, Electrical 4.4, Mechanical 5 | 3.8% | Not reported | Open |
| 100 | Gemma 4 26B A4BGooglepipi harness · ±0.2 across runs · $0.29 per task · 84 min per task · Structural 0.2, Civil 3.3, Electrical 4.6, Mechanical 4.4 | 3.8% | Not reported | Open |
| 101 | GPT-OSS 120BOpenAIcodexcodex harness (native) · ±0.5 across runs · $0.18 per task · 45 min per task · Structural 1.5, Civil 1.3, Electrical 3.5, Mechanical 5.7 | 3.8% | Not reported | Open |
| 102 | GPT-OSS 120BOpenAIhermeshermes harness · ±0.1 across runs · $0.16 per task · 48 min per task · Structural 1.8, Civil 2.4, Electrical 4.9, Mechanical 3.7 | 3.7% | Not reported | Open |
| 103 | GPT-OSS 120BOpenAIopencodeopencode harness · ±0.7 across runs · $0.16 per task · 47 min per task · Structural 1, Civil 2, Electrical 5.4, Mechanical 3.6 | 3.7% | Not reported | Open |
| 104 | MiniMax M2MiniMaxagyagy harness · ±0.2 across runs · about $0.33 per task · 13 min per task · Structural 0.3, Civil 1.8, Electrical 5.7, Mechanical 3.4 | 3.5% | Not reported | Unknown |
| 105 | Kimi K2 ThinkingMoonshot AIagyagy harness · ±0.7 across runs · about $0.29 per task · 18 min per task · Structural 0.1, Civil 1.5, Electrical 3.8, Mechanical 4.2 | 3.2% | Not reported | Unknown |
| 106 | Kimi K2 ThinkingMoonshot AIcodexcodex harness · ±0.8 across runs · $0.26 per task · 9 min per task · Structural 1, Civil 1.3, Electrical 4.4, Mechanical 3.7 | 3.2% | Not reported | Unknown |
| 107 | Gemma 4 26B A4BGoogleopencodeopencode harness · ±0.3 across runs · $0.45 per task · 46 min per task · Structural 0.1, Civil 1.3, Electrical 3.2, Mechanical 3.6 | 2.7% | Not reported | Open |
| 108 | Qwen3-Next 80BAlibabapipi harness · ±0.2 across runs · $1.05 per task (source footnote) · 59 min per task · Structural 0, Civil 0.8, Electrical 5.8, Mechanical 1.7 | 2.7% | Not reported | Unknown |
| 109 | Qwen3-Next 80BAlibabaclaude-codeclaude-code harness · ±0.6 across runs · $0.54 per task (source footnote) · 44 min per task · Structural 1.1, Civil 2.1, Electrical 4.1, Mechanical 2 | 2.6% | Not reported | Unknown |
| 110 | Qwen3-Next 80BAlibabaopencodeopencode harness · ±0.4 across runs · $0.54 per task (source footnote) · 70 min per task · Structural 0, Civil 1.6, Electrical 4.8, Mechanical 1.9 | 2.6% | Not reported | Unknown |
| 111 | Qwen3-Next 80BAlibabahermeshermes harness · ±0.6 across runs · $0.41 per task (source footnote) · 47 min per task · Structural 0.1, Civil 1.4, Electrical 2.6, Mechanical 2.1 | 1.9% | Not reported | Unknown |
| 112 | Gemma 4 26B A4BGooglehermeshermes harness · ±0.2 across runs · $0.29 per task · 24 min per task · Structural 0, Civil 2.6, Electrical 2.7, Mechanical 1.3 | 1.8% | Not reported | Open |
| 113 | Qwen3-Next 80BAlibabaagyagy harness · ±0.6 across runs · about $0.4 per task (source footnote) · 41 min per task · Structural 0, Civil 1.5, Electrical 3, Mechanical 0.6 | 1.4% | Not reported | Unknown |
| 114 | Qwen3-Next 80BAlibabacodexcodex harness · ±0.5 across runs · $0.56 per task (source footnote) · 36 min per task · Structural 0.7, Civil 0.5, Electrical 1.7, Mechanical 0.6 | 1.0% | Not reported | Unknown |
| 115 | Gemma 4 26B A4BGooglecodexcodex harness · ±0.2 across runs · $0.2 per task · 24 min per task · Structural 0, Civil 1, Electrical 0.5, Mechanical 1 | 0.7% | Not reported | Open |
| 116 | Llama 4 MaverickMetaopenhandsopenhands harness · ±0.2 across runs · $0.18 per task (source footnote) · 2 min per task · Structural 0, Civil 0.2, Electrical 1, Mechanical 0.9 | 0.7% | Not reported | Open |
| 117 | Llama 4 MaverickMetapipi harness · ±0.3 across runs · $0.34 per task (source footnote) · 1 min per task · Structural 0, Civil 0.1, Electrical 0.6, Mechanical 1.2 | 0.7% | Not reported | Open |
| 118 | Llama 4 MaverickMetaclaude-codeclaude-code harness · ±0.1 across runs · $0.063 per task (source footnote) · 1 min per task · Structural 0, Civil 0.1, Electrical 0.4, Mechanical 0.6 | 0.4% | Not reported | Open |
| 119 | Llama 4 MaverickMetaagyagy harness · ±0.1 across runs · about $0.06 per task (source footnote) · 3 min per task · Structural 0, Civil 0, Electrical 0, Mechanical 0.2 | 0.1% | Not reported | Open |
| 120 | Llama 4 MaverickMetacodexcodex harness · ±0.1 across runs · $0.097 per task (source footnote) · 2 min per task · Structural 0, Civil 0, Electrical 0, Mechanical 0.1 | 0.0% | Not reported | Open |
| 121 | Llama 4 MaverickMetahermeshermes harness · ±0 across runs · $0.046 per task (source footnote) · 1 min per task · Structural 0, Civil 0, Electrical 0, Mechanical 0.1 | 0.0% | Not reported | Open |
| 122 | Llama 4 MaverickMetaopencodeopencode harness · ±0 across runs · $0.004 per task (source footnote) · <1 min per task · Structural 0, Civil 0, Electrical 0, Mechanical 0 | 0.0% | Not reported | Open |
How we show PLE
We captured Google's Physical-Engineering's Last Exam leaderboard on November 12, 2026. It lists 18 models, most run in seven agent harnesses and some in their own, for 122 model and harness pairs. Each row keeps its harness, the spread across three runs, the cost and time per task, and the score in each of the four engineering disciplines. We did not rerun them.
Scores cover the 56 hidden test tasks (6 structural, 9 civil, 18 electrical, 23 mechanical). Google publishes them from 0 to 1; we show them as percentages. Automated checks grade quantitative outputs, and an LLM judge (Claude Opus 5.5) grades open-ended work against an expert rubric. A design must pass every safety and feasibility check before it earns points for cost.
PLE is display only. Each row is a model inside a specific agent harness, so the table compares complete agent setups, not base models, and it does not enter weighted rankings. A model's native harness is marked. Mercor's copy of this board lists only the native-harness rows.
Snapshot
122 pairs18 models56 hidden tasksDisplay only
The published PLE snapshot places Claude Opus 5.5 first at 48.6%. The third row is 0.9 points behind. The broader top-10 range is 3.7 points, so many of the published results sit in a relatively narrow band.
122 model and harness pairs are shown for PLE. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. PLE is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About PLE
Year
2026
Tasks
56 hidden test tasks (85 in all)
Format
Rubric and check score, 0 to 100
Difficulty
Long-horizon professional engineering
PLE has 85 tasks across structural, civil, electrical and mechanical engineering, built by practicing engineers. Google's leaderboard scores 56 hidden test tasks, three runs per model and harness pair, in an isolated container with no internet and a 6-hour limit. Automated scripts check quantitative outputs and an LLM judge grades open-ended work against an expert rubric. BenchLM mirrors every model and harness pair and keeps it display only, because each row is a complete agent setup.
Freshness and provenance
Version
PLE 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
CurrentDisplay only
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does PLE measure?
Long, end-to-end engineering projects: an agent works through a full project folder, finds the governing codes and physical limits, and produces deliverables a licensed engineer would sign off on.
Which model leads the published PLE snapshot?
Claude Opus 5.5 currently leads the published PLE snapshot with 48.6% score (hidden test set). BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on PLE?
The November 12, 2026 capture snapshot contains 18 models on 9 harnesses.
Last updated: November 12, 2026 capture · mirrored from the public benchmark leaderboard
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.