Introducing SUTRA-R0

Feb 5, 2025

A reasoning model that delivers deeper, structured thinking across topics and domains. Building on our earlier advances, SUTRA-R0 brings complex decision-making, supports multiple languages, and uses resources efficiently—making it valuable for consumer and enterprise use cases.

Update- SUTRA-R0-Preview leads multilingual MMLU benchmark beating DeepSeek-R1-32B and OpenAI-o1-mini in Hindi, Gujarati, Tamil and more.

The release of DeepSeek's R1 model marked a pivotal moment in the AI landscape . While OpenAI’s o1 model set a new course for inference-focused AI reasoning models, DeepSeek-R1 leveraged Reinforcement Learning (RL) and advanced Distillation techniques to create open-weight models that are more usable and cost-efficient. By distilling larger, more complex networks into smaller, optimized models, DeepSeek enabled R1 to tackle intricate problems in unprecedented ways.

At TWO AI, we are keen to introduce our first reasoning model that pushes the boundaries of AI. Building on the breakthroughs of DeepSeek R1, among others, SUTRA-R0 integrates a highly structured reasoning framework. This model deepens the focus on context, language, and complex decision-making. We believe that advanced reasoning transcends mere pattern recognition—it's about truly understanding the underlying logic of situations, a capability that SUTRA-R0 is engineered to deliver.

From Inspiration to Innovation: SUTRA-R0

SUTRA-R0 builds on the core strengths of other reasoning models model to achieve high performance while maintaining resource efficiency. What sets SUTRA-R0 apart is its logical reasoning layer and multilingual capabilities. Rather than merely responding to inputs, it excels at interpreting complex scenarios, solving multi-step problems, and drawing nuanced conclusions across diverse fields.

While R1 specializes in natural language reasoning, SUTRA-R0 employs structured reasoning for deeper analysis. This advancement allows SUTRA-R0 to transcend simple question-answering—it generates practical insights for businesses, uncovers intricate data relationships, and proposes solutions that align perfectly with organizational goals.

SUTRA-R0 leads multilingual MMLU benchmark

SUTRA-R0 not only introduces advanced reasoning capabilities but also retains and, in some cases, surpasses the exceptional multilingual performance established by our previous model, SUTRA-V1. Benchmarking against leading models like Deepseek-R1, OpenAI-o1-mini, and Llama 3.3 70B, SUTRA-R0 demonstrates superior reasoning in Hindi (81.44) and Gujarati (79.39), while maintaining strong performance across multiple languages, including Tamil (77.82) and Bengali (78.91).

SUTRA-R0-Preview is still under active development and is showing significant promise, closing the gap in multilingual MMLU performance to larger models like DeepSeek-R1-600B and OpenAI-o1. The benchmark numbers are reported from the February 3, 2025, checkpoint. Our evaluation methodology follows a rigorous multilingual assessment framework inspired by Lai et al. [2023] and Üstün et al. [2024], with key distinctions. Unlike Okapi’s 25-shot evaluation, we adopt a more challenging 5-shot approach, aligning with Hendrycks et al. [2021]. To ensure broad linguistic coverage, we assess performance across three major language groups: [English] [Korean, Japanese, Arabic ] and [Indian languages]. While not exhaustive, this selection represents significant linguistic diversity, covering over half of the global population and providing a robust measure of SUTRA-R0’s multilingual reasoning capabilities.

In the coming weeks, we will publish a detailed benchmarking report with expanded language coverage, incorporating both quantitative results and qualitative examples for a deeper evaluation.

Unlike other models that often compromise multilingual understanding for reasoning, SUTRA-R0 achieves a balanced integration of both, ensuring precise decision-making across diverse linguistic landscapes. This makes it an invaluable tool for enterprises and consumers who require nuanced, multilingual AI-driven insights.

Enterprise and Consumer Applications

SUTRA-R0 demonstrates promising potential across enterprise applications. In financial services, it enhances risk assessment and market trend analysis. Healthcare organizations benefit from its ability to analyze complex medical data and support research interpretation. In customer service, its structured reasoning framework streamlines inquiry processing and provides intelligent response recommendations.

For consumers, SUTRA-R0's reasoning capabilities deliver personalized experiences in e-commerce, entertainment, and personal finance—helping users make better decisions. It provides tailored product recommendations, optimizes content suggestions, and helps manage personal budgets with relevant, actionable insights.

Initial testing shows SUTRA-R0 effectively handles industry-specific reasoning tasks while complementing existing business processes and improving user experiences. We continue to explore and validate these use cases across enterprise and consumer sectors.

SUTRA-R0-Preview is now available in ChatSUTRA

Try out SUTRA-R0's advanced reasoning capabilities today in ChatSUTRA. Our intuitive interface makes it easy to engage with the model as it thinks through and understands, interprets, and solves multi-step problems. Start exploring the future of AI reasoning at chat.two.ai.

As we continue to refine and advance SUTRA-R0, we're excited about its potential to transform enterprises and industries that demand sophisticated contextual understanding. From real-time decision-making to automated strategy development and data-driven insights, SUTRA-R0 will be a powerful tool for advanced AI applications.

Reasoning models herald a new era of machine reasoning—one where AI systems don't merely process information but understand and apply it in meaningful, actionable ways.

Research @ TWO

0:00/1:34

PUBLISHED

Feb 5, 2025

PUBLISHED

Feb 5, 2025

PUBLISHED

Feb 5, 2025

CATEGORY

Blog

CATEGORY

Blog

CATEGORY

Blog

READING TIME

5 Minutes

READING TIME

5 Minutes

READING TIME

5 Minutes

Recent Posts

Phase 2 of AI in Enterprises

Phase 2 of AI in Enterprises

Enterprises are shifting from AI experiments and demos to fully integrated, scalable solutions that transform core business operations. This transition is redefining industries, creating new business models, and driving innovation across sectors.

Enterprises are shifting from AI experiments and demos to fully integrated, scalable solutions that transform core business operations. This transition is redefining industries, creating new business models, and driving innovation across sectors.

SUTRA-OP: On-Premise Enterprise AI

SUTRA-OP: On-Premise Enterprise AI

On-premise enterprise AI solution designed to meet the data security, privacy, and compliance needs of organizations. With customizable, scalable configurations, SUTRA-OP integrates seamlessly into existing systems while delivering high-performance AI capabilities.

On-premise enterprise AI solution designed to meet the data security, privacy, and compliance needs of organizations. With customizable, scalable configurations, SUTRA-OP integrates seamlessly into existing systems while delivering high-performance AI capabilities.

Introducing SUTRA HiFi

Introducing SUTRA HiFi

A next-generation multilingual voice model that delivers exceptional voice clarity and performance, supporting over 30 languages with ultra-low latency. SUTRA-HiFi’s innovative design and human-like voice quality make it ideal for real-time, global communication.

A next-generation multilingual voice model that delivers exceptional voice clarity and performance, supporting over 30 languages with ultra-low latency. SUTRA-HiFi’s innovative design and human-like voice quality make it ideal for real-time, global communication.