I Said The Rise Of Xiaomi And It Did Happened. Now The Next Big Breakthrough Is Coming From Poolside.

I am betting on Poolside now

In my two previous posts I said the next big step will not come from OpenAI. I said Xiaomi will do it. And Xiaomi did it with MiMo V2.6.

Now I am making the next bet. Same playbook, different company.

Poolside.

Not because their next model will beat Opus 5.5 or Gemini Carbon on day one. It will not. Those are frontier closed models at 88% level.

But because their next model will be a very big breakthrough inside the Poolside family. And I think I know why.

What Laguna S2.1 actually is

Poolside released Laguna S 2.1 on July 21, 2026. It is a 118B total, 8B active per token Mixture of Experts model.

It scores 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE, making it one of the strongest coding models in its category. It has a 1M token context window on the paid endpoint.

It is open weight under OpenMDW-1.1. You can find it on Hugging Face, Baseten, Vercel AI Gateway and OpenRouter. OpenRouter gives a free endpoint with 256K context and a paid endpoint with full 1M context.

The free IDs poolside/laguna-s-2.1:free were live for a very long time, until the expiration date of Oct 31, 2026. That is months of free use.

And people used it a lot. 182B tokens in last 30 days just on OpenRouter for Laguna S 2.1. That is not small.

The free part is not charity. It is data.

This is the same thing Xiaomi did with MiMo. You give the model for free inside the tools where devs live every day. Like OpenCode, Cline, Kilo Code.

When devs use Laguna S 2.1 to fix a repo, run terminal commands, debug a service, every session is an agentic session. It is not a simple chat. It is long, with tool calls, file edits, test runs.

Poolside already trains like this. Their agentic training phase covered 409,000 environments, including 83,000 for terminal tasks and 168,000 for software engineering workflows. The largest single source was about 38,000 real commits from roughly 17,000 repositories.

They also have Model Factory that automates training, eval and experiments. So they can ship fast.

Now add months of free usage on OpenRouter and Cline on top of that lab data. That is pure agentic feedback. Where the model got stuck, how devs fixed it, what final code passed.

Xiaomi got RL gold from free tier. Poolside is getting the same but for coding agents.

Why the next model will be a big jump for Poolside

Let me be clear. I am not saying Poolside next model will be the new frontier king. Opus 5.5 and Gemini Carbon are still ahead.

Laguna S 2.1 is at 70.2% on Terminal-Bench. Frontier is at 88%. That gap is real.

But inside Poolside line, the growth is fast.

Laguna XS.2 was 33B total. Then XS 2.1 added native reasoning. Then S 2.1 jumped to 118B and beat models 10x bigger. Each step was a clear step up.

The next model will have:

1. More real agentic data than any open model. Because free for months.
2. More experience from building M.1, XS.2, XS 2.1, S 2.1.
3. A fast factory to test new ideas.

So it will not be Opus 5.5 level. But it will be the biggest step Poolside has ever made. It may be the model that finally puts an open weight model close to the top 10 on Terminal-Bench and SWE-Bench.

That is why I am betting on it. Same as Xiaomi. Free distribution now, big model later.

Watch Poolside, not just Anthropic.