You got me beat a little bit lol

Server build

We swapped Kimi K3.1 Pro into Claude Code and ran a tricky billing-date task 5 times against Sonnet 5.5. Both passed every hidden test, 5 of 5, but Kimi cost $0.127 a run vs $0.054 and took 148s vs 18s. codetwig.dev/recipes/claude-…

For those who are testing local models, what’s your setup?

Does adding "think hard" help Claude Code on tricky logic? We ran a billing-date function full of edge cases 6 times. Plain prompt or think-hard, all 6 passed every hidden test. The real difference: only the think-hard runs wrote and ran their own tests (3 of 3 vs 0 of 3), at $0.091 vs $0.057 a run.

We asked Claude Code to build a tricky billing-date function 6 times. Plain prompt or "think hard", all 6 passed every hidden edge-case test. Only the think-hard runs wrote and ran tests: 3 of 3 vs 0 of 3, for $0.091 vs $0.057 a run. codetwig.dev/recipes/think-h…

We gave Claude Code the same 500 bug six times. Describing it or pasting the stack trace, all 6 fixed the root cause, but only pasted-error runs also fixed the same crash in tag counts: 2 of 3 vs 0 of 3, for $0.049 vs $0.044 a run. codetwig.dev/recipes/paste-t…

We ran Qwen3.8-27B on two home AMD R9700s against Claude on 6 auto-graded tasks. With thinking on, Qwen passed 15 of 18 for $0 in API; Sonnet 5.5 passed 18 of 18 for $0.30. Downside: 5 of Qwen's 6 diffs failed git apply. codetwig.dev/recipes/qwen-27…

We had Claude Code add a feature to a 24-file API 6 times, half of them in plan mode first. All 6 runs passed our hidden tests either way, but planning first cost 73% more per run: $0.28 vs $0.16. codetwig.dev/recipes/plan-mo…

We asked Claude Code to split a bill 6 times. Asked plainly, 1 run in 3 quietly overcharged someone 2 cents. Add one sentence, "write failing tests first", and all 3 runs got it right, for about 2¢ more per run. codetwig.dev/recipes/tests-f…