PERSONAL CODING AGENT BENCHMARK
Capy Bench Leaderboard
Evaluating coding harnesses and language models on an end-to-end, backend-focused CLI engineering plan (S1âS7) in Elixir/OTP. Tracking completion, duration, token usage, cache efficiency, and real cost.
Personal Code EvaluationBackend & CLIIndependent, hands-on runs on a real backend/CLI workflow (HTTP/TLS, concurrency, timeouts, and signal handling) â meant as practical field notes rather than a formal academic benchmark.
- Total Runs
- 9
- Harnesses
- 4
- Fastest Run
- 25m
- Lowest Cost
- $0.87
| Rank | Coding Harness & Model | Details | |||||
|---|---|---|---|---|---|---|---|
| #1 | ZCodeglm-5.3-flash | 7/7 stages (100%) | 55/55 passed 7 commits | 1h 11m 413 calls | 218.7k in / 147.2k out Cache: 99.15%Ctx: 202.9k peak (26,844 â 202,909 tokens) | $0.87 | Inspect ZCode glm-5.3-flash |
| #2 | Pigpt-6.1-sol | 7/7 stages (100%) | 21/21 passed 8 commits | 40m 358 calls | 144.1k in / 44.9k out Cache: 97.77%Ctx: 75.0k peak (2,854 â 75,009 prompt tokens; 75,070 total) | $1.37 | Inspect Pi gpt-6.1-sol |
| #3 | Pigpt-6-sol | 7/7 stages (100%) | 17/17 passed 8 commits | 27m 325 calls | 96.8k in / 41.9k out Cache: 98.28%Ctx: 74.4k peak (1,753 â 74,400 tokens) | $1.72 | Inspect Pi gpt-6-sol |
| #4 | Codex CLIgpt-6-sol | 7/7 stages (100%) | 31/31 passed 8 commits | 25m 298 calls | 151.1k in / 63.2k out Cache: 98.65%Ctx: 74.9k avg (~15k â ~135k peak tokens across 149 calls) | $3.14 | Inspect Codex CLI gpt-6-sol |
| #5 | antigravity cligemini-3.8-flash-high | 7/7 stages (100%) | 41/41 passed 7 commits | 46m 775 calls | 1.05M in / 151.9k out Cache: 98.11%Ctx: 26.88% out of 1.0M (268.8k peak) | $5.43 | Inspect antigravity cli gemini-3.8-flash-high |
| #6 | ZCodeglm-5.3 | 7/7 stages (100%) | 80/80 passed 8 commits | 1h 36m 532 calls | 259.5k in / 204.6k out Cache: 99.42%Ctx: 271.4k peak (26,846 â 271,444 tokens) | $12.90 | Inspect ZCode glm-5.3 |
| #7 | ZCodeglm-5.3-flash | 6/7 stages (86%) | 42/42 passed 8 commits | 1h 51m 708 calls | 164.6k in / 103.0k out Cache: 99.52%Ctx: 167.7k peak (26,398 â 167,669 tokens) | $1.10 | Inspect ZCode glm-5.3-flash |
| #8 | ZCodeglm-5.3-flash | 6/7 stages (86%) | 63/63 passed 9 commits | 1h 50m 589 calls | 317.7k in / 221.2k out Cache: 99.39%Ctx: 302.3k peak (26,838 â 302,319 tokens) | $1.70 | Inspect ZCode glm-5.3-flash |
| #9 | ZCodeglm-5.3 | 2/7 stages (29%) | 36/37 passed 8 commits | 1h 33m 609 calls | 518.6k in / 115.8k out Cache: 98.44%Ctx: 189.5k peak (26,945 â 189,505 tokens) | $9.74 | Inspect ZCode glm-5.3 |
| No benchmark runs match the current filters. | |||||||