Consumer RTX 5090 code reviewer scores F1 22.7 on Martian benchmark
Obvious-Fan-3183 · reddit · 2026-07-22
Built a code-review system running entirely on a consumer RTX 5090 with open-weight models and no paid APIs.
- Reported P 19.2 / R 27.7 / F1 22.7 on the Martian code-review benchmark.
- The author says the pipeline fits on a single consumer GPU and has already been benchmarked multiple times.
- They’re now asking whether it’s strong enough to ship as a product or open-source.
- The post also suggests there may still be room to push the score toward F1 30, and possibly higher with paid APIs.
More from coding & agent
- Obsidian’s 5 official agent skills teach Claude native notes, Bases and JSON Canvas — FinanceYF5 · 2026-07-22
- One shared vault lets every Claude Code project reuse the same knowledge base — FinanceYF5 · 2026-07-22
- Six daily Claude commands power the Obsidian second-brain workflow — FinanceYF5 · 2026-07-22
- Paper and shadcn make agentic UI block generation feel almost automatic — shadcn · 2026-07-22
- Grok Build adds token usage, batching and diagnostics for developers — elonmusk · 2026-07-22
- SWE-Bench Pro has only 11 repos, and that may be too narrow for real SWE work — ZainHasan6 · 2026-07-22