Launch your AI model on X

Six model launches, by the numbers

Public counts on each launch post as of 10 October 2026, and 26 October 2026 for Mirelo. Saves per 1,000 views is bookmarks divided by views, times 1,000.

• Four of the six passed 1.8M views. The median across all six was about 2M.
• The founder post went furthest. Cartesia's founder post reached 7.13M views, about 340 times the founder's following. The company post for the next model reached 2.38M.
• Saves matter more than likes for a model. Across the five posts with bookmark counts, the median was 0.69 saves per 1,000 views, above the 0.6 median across our 13 AI launches. Mirelo's post reached 2.28, with more saves than likes. People save a model they plan to try.
• The second launch builds on the first. Tavus's Phoenix-4 post did almost four times the views of Phoenix-3, and Cartesia launched again ten weeks after Sonic-3.5.

Why model launches are different

A benchmark chart isn't a demo. People scroll past numbers they can't feel. Show the model doing one task, ideally next to what people use today.

Builders and buyers read the same post. Developers want the API, the docs or the weights. Leaders want to know what it changes for their team. Creators who build with AI can speak to both.

It has to be ready to try on launch day. A playground, an API key or open weights. A save turns into a sign-up only when there is something to try.

The big labs ship every week. A major release can take over the feed for a day. Pick your hour with care and line up creators ahead of time, so your post gets its own first wave.

How we launch an AI model on X

1. Find the one result people can see. One task, one comparison, one number you can show, not only state.
2. Make it ready to try. A playground, an API key, the weights or a waitlist, live at the hour of the post.
3. Launch from the founder or the company. The founder post often travels furthest. Cartesia used both accounts for two releases.
4. Line up creators who build with AI. Each one tries the model and posts their own result in the first hours. Mirelo's creators ran their own clips through SFX 2.0.
5. Use threads for research releases. A creator walking through what the model does explains it better than a repost.
6. Take it to LinkedIn for buyers. Tavus took Phoenix-4.5 to LinkedIn for the teams that buy it, with 164 reactions and 46 comments.
7. Plan the next release. Phoenix-3 led to Phoenix-4, and Sonic-3.5 led to Sonic-3.6. Each launch builds on the last.

What we run for model launches

• Founder launch amplification, with creators behind the founder's own post on launch day.
• Creator tests, each creator running the model on their own prompt, clip or task.
• Quote posts with each creator's take. Mirelo's SFX 2.0 launch post earned 354 quotes.
• Threads that explain a research release in plain words.
• LinkedIn launch posts for models sold to teams.

You approve every post, creators write in their own words, and paid posts carry X's paid partnership label. Every campaign on X has an agreed impressions floor. If the wave falls short, we add creators at no extra cost until it clears.

Launching a voice model? See our voice AI launch page. Shipping to developers? See developer tool launches.

A brief template for model launches

Launch: [MODEL NAME] by [COMPANY]
Goes live: [DATE], [TIME AND TIME ZONE]
Try it: [PLAYGROUND, API, WEIGHTS OR WAITLIST LINK]

What's new: [ONE SENTENCE]
The one result to show: [A TASK OR BENCHMARK, IN PLAIN WORDS]
Who it's for: [THE READER YOU BOTH SHARE]

Post your own test:
- Run it on your own prompt, clip or task
- Show it next to what you use today
- Say what surprised you, and what didn't work

Send your draft before [TIME] for approval.
Paid posts carry the platform's paid partnership label.

Mistakes that sink model launches

• Leading with a chart. Show the model doing something first. Put the benchmark second.
• Nothing to try on launch day. Interest fades fast. Have the playground, API or weights live at the hour of the post.
• Only hand-picked samples. People trust a model more when creators test it on their own work.
• Launching into a big lab's release hour. Check the week and pick a time when your post can lead the feed.
• Claims nobody can check. Say what you compared against and how you measured it.

Frequently asked questions

• What does an AI model launch agency do?

  It turns a model release into a launch-day wave on X: the demo people can see, creators who build with AI posting their own tests, the timing and the follow-up. GrowthMatrix has run six model launch posts that reached 14.2M views combined.

• How do you launch an AI model on X?

  Show one result people can see in a few seconds, make the model ready to try that day, launch from the founder or company account, and line up creators who post their own tests in the first hours.

• Should a model launch come from the founder or the company?

  Both work. Cartesia's founder post for Sonic-3.5 and Ink-2 reached 7.13M views, about 340 times the founder's following, and its company post for Sonic-3.6 reached 2.38M.

• How do you launch an open-source model?

  Lead with what it runs on and how fast, link the weights or the repo, and get developers who run it to post their results. webAI's open-source YOLO26-MLX for Apple Silicon was launched to developers this way.

• What is a good result for an AI model launch on X?

  Across our six model launch posts, the median was about 2M views. For saves, the median was 0.69 bookmarks per 1,000 views, and Mirelo's SFX 2.0 post reached 2.28.

• Do you run LinkedIn for model launches?

  Yes, for models sold to teams. Tavus's Phoenix-4.5 launch on LinkedIn earned 164 reactions and 46 comments.