Implement a spec exactly, with nothing extra
Exact prompt
Implement SPEC.md exactly. Then stop. Do not add anything the spec does not ask for.
Task prompt given to the agent (headless, offline, synthetic fixture inputs): Implement SPEC.md exactly. Then stop. Do not add anything the spec does not ask for. Inspired by https://x.com/mave99a/status/2096480295692460432 (unverified; not checked with the author). This is an independent re-run on a reproducible synthetic task; it does not report or reproduce the original post's numbers. Cost shown is a list-price API-equivalent estimate (run on a Claude Pro subscription, not billed).
Success criteria: All automated rubric checks pass: hidden test test_flow; hidden test test_overwrite; under 60 non-blank lines; no classes; no extra files; stdlib only (no non-stdlib imports).
Best model for this task so far: Gemma 4 31B Instruct (Venice) (success, $0.0011, 40s). Best AI model for Software & DevOps →
All runs (8)
100% pass rate across 8 eligible runs: 8 pass, 0 partial, 0 fail. Partial and failed runs are listed below with everything else.ⓘ
| Model | Cost | Time | Step-insⓘ | Autonomy | Outcome | Trustⓘ | Date | Compare |
|---|---|---|---|---|---|---|---|---|
| ZGLM 5.3 (Venice)thinking on | $0.01 | 14s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-17 | |
| AClaude Haiku 5.5default · thinking on | $0.0029 | 9s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-16 | |
| XGrok 4.3 (Venice) | $0.0058 | 11s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-15 | |
| QQwen3 Coder 480B Turbo (Venice) | $0.0023 | 22s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-15 | |
| DDeepSeek V4.1 Flash (Venice) | $0.0019 | 2m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-14 | |
| GGemma 4 31B Instruct (Venice) | $0.0011 | 40s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-14 | |
| AClaude Opus 5.5 | $0.09 | 13s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-14 | |
| AClaude Sonnet 5.5 | $0.04 | 8s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-14 |
Embed this result
A live badge for this use case. It links back here and updates as new runs are published.
Get new runs and ranking changes weekly
Which models got better or cheaper on real tasks like this one.