The $25 Challenge

What can an agent build when the meter stops at $25?

Rules

Set your cap before you start: gremlord budget set --daily 25. That is a daily cap, so it does not by itself keep a multi-day build under $25. The contest receipt is gremlord cost --receipt --since YYYY-MM-DD from the day you started the build — the from/to lines on the paste are the window. That number plus any local hardware cost (rule 5) must be ≤ $25.00. Bare gremlord cost --receipt (no --since) is today's spend, not the contest total, and a lifetime dump is not an entry.
Build anything real: a tool, a fix, a feature, a demo. The entry must perform a function beyond presenting itself.
Enter by replying to the pinned "The $25 Challenge" discussion with: what you built (one paragraph), a link to the repo or demo, and your gremlord cost --receipt --since YYYY-MM-DD output pasted in full as the receipt.
Used a passthrough profile or subscription-billed CLI delegation (Codex/Grok under your own ChatGPT/SuperGrok login) for any part of the build? Those runs show $0 or don't appear in gremlord cost at all. Disclose it in your entry. Unpriced subscription work is recognition-only — it cannot win the $25 bounty unless you assign it a fixed rate and include that rate in the contest total.

Used a local/self-hosted model (Ollama, vLLM, etc.)? Those don't have a metered API price either, so gremlord cost won't show real spend for them. Price them by fixed hardware rate × active runtime instead, and add it to your tracked total.

Active runtime is from request start / model load through the final token. Sum every run. Round each run up to one minute. Rates assume 3-year hardware life at 50% use and $0.20/kWh.

Tier	Hardware	Rate
1	Laptop / mini PC, base Apple M-series, GPU ≤12GB	$0.10/hr
2	High-memory workstation, M Pro/Max, Ryzen AI Max	$0.25/hr
3	One 16–24GB consumer GPU + host	$0.35/hr
4	Two consumer GPUs or one 48GB pro GPU + host	$0.70/hr
5	Datacenter accelerator	actual rental rate, else $3/hr

Match your model to a tier: 4–12B Q4 → tier 1. gpt-oss 20B or Qwen3-30B-A3B → tier 2 or a 24GB GPU (tier 3). Dense 32B → tier 2 or tier 3. 70B Q4 → tier 4 (64GB+ unified memory or two GPUs). Submit your gremlord receipt, your hardware tier, and your active runtime alongside it — tier rate × runtime counts toward the $25 cap.

Honor system on receipts. A faked log to win a README mention is a sad life.
One entry per person. Entries close September 17, 2026 23:59:59 UTC.

How to enter

Install gremlord

One command, see the README for details.

Set your $25 cap

gremlord budget set --daily 25 — the router refuses the next request once you hit it.

Print your receipt

Build, then print your receipt from the day you started: gremlord cost --receipt --since YYYY-MM-DD. That number plus local hardware cost must be ≤ $25.00.

Reply with your build + receipt

Reply to the pinned Discussion with your build, that receipt pasted in full, a disclosure line for any passthrough/subscription CLI delegation (and a rate if you want it to count), and your hardware tier + active runtime for any local models.

# install
❯ curl -fsSL https://raw.githubusercontent.com/gremlord/gremlord/main/install.sh | sh
❯ gremlord setup

# cap it at $25/day
❯ gremlord budget set --daily 25

# receipt from the day you started — paste this in full
❯ gremlord cost --receipt --since YYYY-MM-DD

Judging & prize

Community pick

Most thumbs-up on the discussion entry at cutoff. Founder breaks ties.
Recognition only — permanent leaderboard + README slot.

Founder's pick

Most impressive build under budget, chosen by @MaorBril.
Wins a $25 bounty, plus a permanent leaderboard + README slot. Unpriced subscription-billed work (rule 4) is recognition-only and cannot take the bounty.

Leaderboard

Standings update daily.

Rank	Builder	Build	Spend	Link
Loading standings…

Entries close September 17, 2026 23:59:59 UTC (7 days after the announcement). Standings also live on the README leaderboard.

Enter before the window closes.

Enter on GitHub →