TuRTLe - Icarus Synthesis: leaderboard
The "Icarus Synthesis" subtask of TuRTLe. A task from TuRTLe, evaluating LLMs on RTL / Verilog hardware-description-language code generation and completion, checked with hardware simulators.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Around January 2027. 45 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 78.65 | #144 |
| 2 | DeepSeek R1 0528 | 75.33 | #222 |
| 3 | GLM-5 FP8 | 75.22 | #143 |
| 4 | DeepSeek R1 | 74.12 | #251 |
| 5 | Gemma 4 31B (IT) | 73.67 | #176 |
| 6 | DeepSeek V3.1 Terminus | 72.28 | #220 |
| 7 | GPT-OSS-120B | 69.45 | #341 |
| 8 | Kimi K2 Instruct (0905) | 69.25 | #257 |
| 9 | Gemini 2.5 Flash (Medium) | 66.32 | #245 (Gemini 2.5 Flash) |
| 10 | Qwen 3 Coder 480B A35B | 63.03 | #279 |
| 11 | GPT-OSS-20B | 59.62 | #518 |
| 12 | Llama 3.1 405B | 53.8 | #466 |
| 13 | QwQ-32B | 52.8 | #421 |
| 14 | Qwen 2.5 72B | 48.94 | #499 |
| 15 | Qwen 2.5 32B | 46.43 | #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-icarus-synthesis · How It Works · Data refreshed daily, snapshot 2026-10-20.