Results
Agents
Drag to turn the plot, or focus it and use the arrow keys. Choose an agent for its values and ranks.
| # | Agent | Average scoreScore | Tokens per taskTokens/task |
|---|---|---|---|
| 1 |  gpt-6-astra | 0.60 | 30.0M |
| 2 |  glm-5.3 | 0.57 | 22.8M |
| 3 |  qwen3.8-max | 0.49 | 24.0M |
| 4 |  claude-opus-5-5 | 0.48 | 11.4M |
| 5 |  kimi-k3 | 0.42 | 12.4M |
| 6 |  gemini-3.1-pro-preview | 0.27 | 6.6M |
| Agent | Synthesis score | Physical-design score | Tokens per task |
|---|---|---|---|
| claude-opus-5-5 | 0.644 | 0.308 | 11.4M |
| gemini-3.1-pro-preview | 0.260 | 0.276 | 6.6M |
| glm-5.3 | 0.603 | 0.529 | 22.8M |
| gpt-6-astra | 0.638 | 0.550 | 30.0M |
| kimi-k3 | 0.452 | 0.393 | 12.4M |
| qwen3.8-max | 0.519 | 0.455 | 24.0M |
Contact
If this draws your attention, and you’re seeking data, benchmarks, evals, and RL envs for Synthesis, Physical Design and such tasks, we should talk — fill in the form below. The above plot indicates a subset of performance on easier tasks. Hard tasks that take 50M tokens and 80 hours are available on request, testing the limits of EDA tool orchestration by current frontier models, with rigorous evaluation tested vs the best possible workflows by industry veterans. Excellent PDK and design coverage across a wide range of flow parameters and difficulty.