Lexa‑Rho: Robi Labs Launches Its First Reasoning‑First Model

At Robi Labs, we believe that the future of AI isn’t just about generating answers — it’s about the ability to reason. Over the last few years, AI models have evolved from completing sentences to tackling complex tasks. But to truly help people, these systems need to think through problems, not just match patterns.

Today, we’re proud to introduce Lexa‑Rho: the first reasoning-first model developed by Robi Labs and the Lexa family. Currently in internal testing and gradually rolling out to more users, Lexa‑Rho marks the beginning of a new era — one where AI doesn’t just respond, but reflects, plans, and solves with intention.

Deep reasoning, not just answers

Traditional models are great at surface-level responses. Ask a factual question, and they’ll likely answer correctly. But when a task requires multi-step logic, deduction, or strategic planning, they fall short.

Recent breakthroughs from other labs have shown why reasoning matters. OpenAI’s o1 series, for example, used internal “reasoning tokens” to enable slower, more deliberate thought — like a mathematician solving a proof rather than blurting out whatever comes first. These models performed better on tough benchmarks because they spent more time thinking.

Lexa‑Rho builds on that insight. Its architecture is designed from the ground up to support multi‑step reasoning. Faced with a challenging problem, Lexa‑Rho breaks it down, explores possible paths, and only then responds.

• Olympiad-style math

• Graduate-level science

• Intricate coding challenges

Benchmark Performance

Reinforcement learning that rewards thinking

To help Lexa‑Rho think more deliberately, our team designed a custom reinforcement learning (RL) objective. Inspired by approaches like DeepSeek‑R1, we train the model not just to get the right answer, but to value the journey — rewarding correct intermediate steps and discouraging shortcuts or fabricated logic.

This RL objective helps reduce hallucinations and encourages logical consistency. Put simply: Lexa‑Rho learns to reason carefully, not just guess quickly.

Smarter conversations with Lexa Chat v2.0

Lexa‑Rho will power the new Lexa Chat v2.0, our next-generation AI interface. Testers will find conversations more coherent, responses more grounded, and tool use more reliable than ever.

• Math & science help: Work through derivations and detailed explanations

• Coding assistance: Spot bugs, explain code flow, and build verified solutions

• Decision support: Explore trade-offs, surface assumptions, and go beyond summaries

Gradual and inclusive rollout

We’re committed to making Lexa‑Rho widely available in a thoughtful way. Rollout will happen in three stages:

1. Internal testing – happening now, focused on refining quality and safety

2. Premium access – launching soon with Lexa Chat v2.0, for Pro, Team, Enterprise, Edu, and Partner plans

3. Free tier – later this year, with reasonable daily usage caps

Built for the future

Lexa‑Rho is just the beginning for reasoning-first models at Robi Labs. It’s the first release in the Lexa reasoning series, and we’re already working on larger and specialized variants — including models fine-tuned for:

• Code synthesis

• Scientific discovery

• Knowledge retrieval

We’re also exploring multimodal reasoning and tool-calling capabilities, inspired by research on large-context models like OpenAI’s o1.

Our team is investing in next-gen training techniques too — from mixtures of experts to hierarchical controllers — to extend reasoning depth while improving efficiency. We’ll be sharing more in an upcoming technical paper.

Join the reasoning revolution

Lexa‑Rho is a major step toward our mission: building AI that helps people — not replaces them. As we bring Lexa‑Rho to more users and more products, we’re excited to see how you’ll use it to think deeper, tackle harder problems, and make better decisions.

The reasoning revolution starts now.
Thanks for being part of it — and stay tuned to the Robi Labs research blog for what’s next.