Vestra Robot 1.1
More capable, in a quarter of the words.
Our new model is more capable than Robot 1.0 and gets to the point: in early internal testing, about 46% more capable with about a quarter of the output.
Vestra Robot 1.1 is the new model behind VestraAI. It answers more questions well than Robot 1.0 did, and it stops when it has answered them. Robot 1.0 had a habit of long, padded replies; Robot 1.1 says the useful part and leaves the rest out.
Vestra Robot 1.1 is our most capable model yet. We’ve sharpened its reasoning, fine-tuned it to hold up under the most demanding work, and engineered it so that it simply works. On our own benchmark suite it does markedly better than our previous model while writing far fewer tokens, with the same context and the same load.
Note
About these numbers
The figures on this page come from early internal testing. They are preliminary and may change as we test further. We’ll update this page if they do.
At a glance
- ~46%More capable than Robot 1.0 on our internal evaluations
- ~4×Fewer output tokens in a typical answer
- ~5–6×Capability per output token, derived from the two estimates
More capable, less to read
On our internal evaluations, Robot 1.1 scores about 46% higher than Robot 1.0. Over the same questions, its answers use about a quarter of the output tokens. Put together, that is roughly five to six times as much capability for each token it writes.
Robot 1.1 and Robot 1.0, early internal estimates
Capability
Length
Per token
Show as a table
| Model | Capability | Length | Per token |
|---|---|---|---|
| Robot 1.1 | 146 | 25% | 5.8× |
| Robot 1.0 | 100 | 100% | 1× |
We measured both models on our own benchmark suite, with the same context length and under the same load. Robot 1.0 was built on Ornith 1.0 9B; Robot 1.1 is fine-tuned from Qwen 3.8 27B, a newer and larger base model.
Same question, two answers
Shorter isn’t the goal on its own. The point is that the useful part arrives first, and there’s less to read past to find it. Both answers below are written out at the same speed, so you can see how much sooner the second one is done.
Illustrative example
My Minecraft server lags every evening when more players join. What should I check first?
Robot 1.0 style
388 tokens
Great question! Server lag can be really frustrating, and there are many possible causes. Before we dive in, it’s important to understand that lag can come from lots of different places, including your server’s hardware, your network, your plugins, and even your players’ own internet connections. Let’s go through some of the most common causes in detail.
- Check your server’s performance. Performance is one of the most important things to look at. You can use tools such as spark or timings to see how your server is performing. These tools can show you which parts of your server are using the most resources, which can help you find the cause.
- Check your RAM. RAM is very important for Minecraft servers. If your server doesn’t have enough RAM, it may lag. However, giving your server too much RAM can also cause problems in some cases, so it’s important to find the right balance for your server.
- Check your plugins and mods. Some plugins and mods can use a lot of resources. Try removing plugins one at a time to see whether the lag improves.
- Check your view distance. A high view distance means the server has to load more chunks, which can cause lag, especially with more players online.
- Check your players’ connections. Sometimes lag is caused by players’ own internet connections rather than by the server itself.
In summary, there are many possible causes of server lag, and the best approach is to investigate each one carefully. I hope this helps! Let me know if you have any other questions, and good luck with your server!
Robot 1.1
91 tokens
Run /spark profiler during the slowdown and check MSPT (milliseconds per tick).
- Over 50 ms: the server can’t keep up. The profile shows why: usually entities, a heavy plugin or a high
simulation-distance(try 6). - Under 50 ms: the lag is network-side, not the server.
If MSPT is high and nothing in the profile stands out, the server needs more CPU than its plan has.
Available now
Robot 1.1 is live today, and it has replaced Robot 1.0, which is now retired.
| Where | Robot 1.1 | What you need to do |
|---|---|---|
| VestraAI | Live now: chat, Kite and voice calls | Nothing. You’re already using it. |
| Vestra AI API | Live now on Vestra Platform | Nothing. Requests now go to Robot 1.1; keys, usage and plans are managed on Vestra Platform. |
| Robot 1.0 | Retired | Nothing. It has been replaced everywhere. |
Kite runs on Robot 1.1
Kite, the personal assistant inside VestraAI, now runs on Robot 1.1 too. It sets reminders, runs recurring tasks such as a weekly news digest, keeps your lists and watches for things so it only messages you when they happen. Shorter, more accurate answers matter most there: Kite’s messages arrive while you’re busy, so they should say what you need and stop.
What hasn’t changed
VestraAI still runs on Vestra’s servers in Germany, and your conversations are saved to your account as before, where you can delete them at any time. There’s nothing to update or switch on: Robot 1.1 is already the model behind VestraAI.
Try it
Robot 1.1 is in VestraAI today, and on Vestra Platform for developers. Questions, and examples where it gets something wrong, are welcome at [email protected].
Figures on this page are early internal estimates. They are preliminary and may change.
