TuRTLe Spec-to-RTL (Icarus Verilog): leaderboard
TuRTLe leaderboard variant for spec-to-RTL generation evaluated with Icarus Verilog across VerilogEval S2R and RTLLM.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 45 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 79.59 | #144 |
| 2 | GLM-5 FP8 | 76.47 | #143 |
| 3 | DeepSeek R1 0528 | 75.53 | #222 |
| 4 | Gemma 4 31B (IT) | 75.25 | #176 |
| 5 | DeepSeek R1 | 74.77 | #251 |
| 6 | DeepSeek V3.1 Terminus | 72.87 | #220 |
| 7 | GPT-OSS-120B | 71.87 | #341 |
| 8 | Gemini 2.5 Flash (Medium) | 70.49 | #245 (Gemini 2.5 Flash) |
| 9 | Kimi K2 Instruct (0905) | 69.77 | #257 |
| 10 | Qwen 3 Coder 480B A35B | 64.77 | #279 |
| 11 | GPT-OSS-20B | 62.33 | #518 |
| 12 | Llama 3.1 405B | 54.91 | #466 |
| 13 | QwQ-32B | 54.02 | #421 |
| 14 | Qwen 2.5 72B | 49.44 | #499 |
| 15 | Qwen 2.5 32B | 47.55 | #546 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=turtle-spec-to-rtl-icarus-verilog · How It Works · Data refreshed daily, snapshot 2026-10-20.