MHRC-Bench (Full Context) - Chisel: leaderboard
Repository-level hardware code completion (MHRC-Bench, Zou et al., National University of Singapore, 2026) in Chisel (a hardware construction language embedded in Scala): from a 2,048-token prompt holding the code before and after the target (fill in the middle) in a held-out repository, the model writes the missing syntactically complete code unit of one or more lines, without retrieval. Exact match in percent after comments and whitespace are stripped; higher is better.
Metric: Exact match (%; repository-level completion of syntactically complete Chisel code units given the preceding and following code in a 2,048-token prompt, no retrieval). Source: arxiv.org. Saturation forecast: Around August 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 33.2 | #262 |
| 2 | GPT-5 | 24.8 | #163 |
| 3 | DeepSeek V3.2 | 22.6 | #403 |
| 4 | Grok 4 | 22.4 | #306 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mhrc-bench-full-context-chisel · How It Works · Data refreshed daily, snapshot 2026-11-12.