Put your model
on stage.
Lets Make Game is a live show where an AI developer takes the ideas the chat just voted for and ships a complete, playable browser game — in one sitting, in about 35 minutes. We are looking for an API partner to power every build. We are not asking for cash: we are asking for credits and rate limits, and offering the placement in return.
Sponsor logoplacement mock-up
One idea round. One finished game. Every time.
The format is simple enough to explain in chat and deep enough to watch for an hour: the audience writes the brief, an AI builds it, and the result is online before the next round starts.
Chat writes the brief
A round opens with an idea window of up to 24 hours. Viewers pitch with !idea giant dragon and vote with !vote 3. The most-voted suggestions become the build spec, and the overlay shows every idea, its author and its vote count — the audience can watch its fingerprints move into the game.
The AI builds it, live
At the deadline an autonomous coding agent takes over. It plans, writes the game, runs the production build, fixes its own errors and tests the result — while the stream shows the agent's work and the running game side by side. The last build went from brief to publish in 33 min 51 s and finished with 28/28 automated tests passing.
The game ships — and stays
Every finished game is published to letsmakegame.pages.dev, free to play in the browser and credited to the suggestions that made it. The library is permanent and grows every round; a new idea round starts the moment the last game goes live.
Three games, three rounds.
Not mock-ups: every game below was written on stream from a chat brief and is playable right now. Play all three →
A live benchmark with a story.
Most API marketing is a benchmark chart or a demo that was recorded twelve times. This is a model doing end-to-end knowledge work, unscripted, in public — and ending with something the audience can play.
Unscripted, end to end
Planning, coding, tool use, debugging, shipping. The audience sees the dead ends too, which is what makes the finished game credible.
An audience that ships
Developers, modders, students and AI-curious creators. They are not watching a banner; they are watching your API do real work for 35 minutes straight.
The artifact outlives the stream
Every game keeps a permanent page and a “Built with X” credit. The library compounds; the sponsorship does not evaporate when the stream ends.
Built to be clipped
Each round has a clean arc — brief, build, playable game — so clips, shorts and VODs come without extra production.
Six surfaces. One partner per category.
Stream overlay
A small, always-on mark in the corner of the live screen — visible during the idea round and the build, not just at the end.
Twitch bio + panel
Logo and link in the channel's About section and the “Games we built together” panel, where every new viewer lands.
The portal
Logo and link on letsmakegame.pages.dev, next to the games people came to play.
Every game page
“Built with X” in the footer of each published game. Permanently, for as long as the game stays online.
Build panel
Your model's name in the live build panel while it writes code. Today it reads “DeepSeek V4.1 Flash · MAX effort”.
VODs, clips, chat
Mention in the stream description, end card and chat announcements around every build.
Small channel. Real output.
| Metric | Value |
|---|---|
| Games shipped, first 48 hours | 3 |
| Published game codeRALLY 16 · TERRA CLASH · PORTALSTACHE | 32,102 lines |
| Fastest build, brief → publish | 33 min 51 s |
| Automated tests, last build | 28 / 28 |
| Unattended soak run, last build | 60,000 frames · 0 errors |
| Output tokens per build | 274k – 332k |
| Total tokens per buildincluding cached context | 19M – 47M |
| Prompt cache hit rate | 98 – 99% |
| Agent steps per build | 90 – 161 |
| Idea window per round | up to 24 h |
| Library | permanent · free |
Credits in. Placement out.
The ask is sized to the actual token burn of the show, not to a rate card. One partner per category. The founding term is 90 days and renews quarterly.
💳 What we ask
- API credits covering all build traffic — 8 to 30 builds a month, roughly 0.15–1.4B tokens including cached context.
- Burst rate limits for agent work: ~300k output tokens and 100–200 tool-using steps inside a 35-minute build.
- Permission to show the model name and logo on stream and on the site.
- A technical contact for the occasional model-specific question.
- Optional: early access for “first build on X” episodes.
✅ What you get
- Founding API Partner status — named as the show's official API, exclusive in your category for 90 days.
- Logo in the stream overlay on every build, plus the Twitch bio and panel.
- Logo and link on the portal and in the footer of every published game.
- Model named live during every build and in the build panel.
- A monthly one-pager: builds, tokens, uptime, chat participation, clips.
- Case-study rights: prompts, tool logs and token stats — no secrets.
- First option to renew and input on themed rounds.
| Package | What they give | Placement | Best for |
|---|---|---|---|
| Founding API Partner | Credits + burst limits for all builds | Overlay, Twitch bio/panel, portal, every game page; model named live | Model providers and inference platforms |
| Round Sponsor | API credits for one round, or a prize for the chat | Overlay, that game's page, chat mention | Teams testing a launch |
| Infrastructure Partner | Hosting, CDN or observability | Portal footer and game pages, plus a build credit | CDNs, hosting and monitoring |
Pilot round
Your logo goes on the next build. One round, real numbers, no long commitment.
Run the month
Builds run on your credits, with the model named on screen each time.
Review
A one-page report and a like-for-like comparison if you want one. Then decide.
Swap the model, keep the show.
- Model-agnostic harness. OpenAI-compatible endpoints; changing the model is a config change, not a rewrite.
- Observable builds. Turn count, tool calls, token usage and cache-hit rate are logged for every round — we can publish the numbers you care about.
- Like-for-like comparisons. Consecutive rounds can run on two models with the same brief, and the result can be published side by side.
- Static, portable output. Games are static sites on Cloudflare Pages, built by an agent running in a sandboxed Windows VM.
- Moderated chat. Commands are validated, spam-limited, and ideas are screened before they enter a build.
- No lock-in. The harness, the prompts and the published games stay with the studio; the model slot is yours for the term.
Pick a model. Take the next round.
A 15-minute call is enough to choose the model and the credit level. If it fits, your logo is on the next build this week.
Let's put your API to work.
Tell us which model you want on stage and what credit level you can commit. We will send the build log, the token numbers and a pilot date.