40% commission on every sale of The 2026 Bench Report ($14.90 main edition). That's $5.96 per sale — or $4.77 on every discounted launch sale — paid automatically through Gumroad. No cap, 60-day cookie.
Gumroad handles everything: tracking, attribution, payouts, and refunds. You just share your link.
We test benchmarks for a living — the affiliate terms get the same treatment.
| Term | Value | Why |
|---|---|---|
| Commission rate | 40% | Upper end of the 30–50% info-product norm; we'd rather overpay early affiliates than beg for them later. |
| Cookie window | 60 days | Longer than Gumroad's 30-day default; buyers of technical reports comparison-shop slowly. |
| Payout | Via Gumroad, PayPal/bank | No minimum threshold; Gumroad's standard payout schedule and fees apply. |
| Refunds | Clawed back | 30-day refund policy on the product; commission follows the money. |
| Discount codes | Stacking allowed | Commission counts post-discount (e.g., LAUNCH20 sales → $4.77/sale). |
| Rules | No brand bidding, no spam | Don't bid on the product name in ads; don't mass-DM. Content-based promotion only. |
Grab and go. Edit freely — in our testing, rewritten first lines convert better than copy-paste.
Someone finally benchmarked AI coding tiers with actual receipts: 85 tasks, 305 logged API calls, every number traceable to a raw run JSON. The cheap tier won. 97.6% vs 94.1% — and the expensive one was slower and burned more tokens. Full data + a decision tree: [YOUR LINK]
Recommended this week: The 2026 Bench Report. One operator ran 85 gradable coding tasks through the same model family at three price tiers and published everything — the runs, the grader's rulebook, even the failures. The uncomfortable answer: the cheap tier scored higher, ran faster, and cost less. Where the small model actually dies (it's not the hard problems) is worth the price alone. → [YOUR LINK] — $14.90, refundable in 30 days.
• "Hard with a known recipe" beats "medium with no shape" for small models — 80-100% vs 20%.
• The scariest bug class: the model patches your code and silently un-patches something you shipped. slugify('Hello World') → 'helloworld'.
• Zero problems across six batches where the expensive tier genuinely won.
• Three prompt rules that erase the tier gap.
I stopped guessing on this after reading an actual logged benchmark — 85 tasks, all runs published. The short version: cheap thinking tier for everything, verification step on small models, and a decision tree for edge cases. The report is The 2026 Bench Report ([YOUR LINK]) — the sample chapters are free if you just want the conclusions.
• The expensive AI tier scored lower (I have the receipts) • 85 tasks, 3 tiers, 1 uncomfortable answer • Your AI subscription's expensive tier is costing you money • Where small models actually die (not where you think)
Ready-made banner set in /banners/ — 1200×628, 728×90, 300×250. Dark charcoal + burnt orange, matches the book cover. PNG export or serve the HTML as-is.
Tell us where you'll promote it. Dev newsletters, AI YouTube, engineering blogs, and niche communities are all fits. We read every application.