schubydoo / clauster
Your homelab's Claude Code launchpad. Self-hosted web UI that spawns & manages remote-control bridges on a remote host from any browser or phone — start/stop, spawn & permission modes, CLAUDE.md editor, git clone, live log tail, cost tracking. Loopback by default; password/reverse-proxy auth. No telemetry.
Languages
Harness | Input / Output Cost | ||||||
|---|---|---|---|---|---|---|---|
1 |  Mini-SWE-agent | 29 / 30 | $0.68 | $2/$12 | 5m02s | ||
2 | Mini-SWE-agent | 29 / 30 | $0.15 | $0.2/$1.2 | 8m23s | ||
3 | Mini-SWE-agent | 29 / 30 | $8.38 | $5/$25 | 22m35s | ||
4 | Mini-SWE-agent | 28 / 30 | $3.12 | $4/$20 | 10m24s | ||
5 | Mini-SWE-agent | 27 / 30 | $0.93 | $2/$6 | 8m37s | ||
6 | Mini-SWE-agent | 27 / 30 | $1.49 | $1.25/$4.25 | 9m04s |
Key Takeaways
- GPT-5.6 Sol with Mini-SWE-agent resolves 28 of 30 tasks, while Grok 4.5 and Muse Spark 1.2 with Mini-SWE-agent each resolve 27.
- Among 29-task results, GPT-5.6 Luna with Mini-SWE-agent costs $0.15 per test, versus $0.68 for GPT-5.6 Terra and $8.38 for Claude Opus 5.
- GPT-5.6 Terra with Mini-SWE-agent records 301.50 seconds average latency, compared with 503.12 for GPT-5.6 Luna and 1354.75 for Claude Opus 5.
Cost Analysis
Average Token Use / Test
No token usage data available.
Cost is the clearest tradeoff in this comparison. Claude Opus 5 leads at 96.67% for $8.38 per test. GPT-5.6 Luna is the lower-cost option at 96.67% for $0.15 per test.
Latency Analysis
Average Response Time / Test
Latency separates several models with similarly strong scores. Claude Opus 5 leads at 96.67%, while GPT-5.6 Terra is fastest at 5m 2s with 96.67% accuracy.