Introducing Sonus-1: A New Era of LLMs

At Sonus AI, we're driven by a mission to push the boundaries of AI. We're thrilled to announce the release of the Sonus-1 family of Large Language Models (LLMs). Sonus-1 has been designed to bring the perfect blend of high performance with versatility across various applications.

At Sonus AI, we're driven by a mission to push the boundaries of AI. We're thrilled to announce the release of the Sonus-1 family of Large Language Models (LLMs). Sonus-1 has been designed to bring the perfect blend of high performance with versatility across various applications.

Meet the Sonus-1 Family: Pro, Air, and Mini

The Sonus-1 series is tailored to meet various needs:

  • Sonus-1 Mini: For when speed is of the essence, offering high-speed and cost-effective solutions

  • Sonus-1 Air: A versatile model with a balance of performance and resource usage.

  • Sonus-1 Pro: Our top-tier model, optimized for complex tasks requiring the best performance.

  • Sonus-1 Pro (w/ Reasoning): Our flagship model featuring chain of thought reasoning for even more capability.

Benchmark Performance

The Sonus-1 family of models demonstrates impressive performance across different areas, with improvements in general reasoning, mathematics, and coding. Here are some highlights:

  • MMLU: Sonus-1 Pro (w/ Reasoning) reaches 90.15% on MMLU, demonstrating its powerful general reasoning skills.

  • MMLU-Pro: Sonus-1 Pro (w/ Reasoning) reaches 73.1% on MMLU-Pro, showing its robust capabilities.

  • Math (MATH-500): Sonus-1 Pro (w/ Reasoning) excels with a score of 91.8%, highlighting its ability to tackle complex math problems.

  • Reasoning (DROP): Sonus-1 Pro (w/ Reasoning) reaches 88.9%, showing its strong capabilities in reasoning tasks.

  • Reasoning (GPQA-Diamond): Sonus-1 Pro (w/ Reasoning) scores 67.3% on the challenging GPQA-Diamond benchmark, emphasizing its aptitude in scientific reasoning.

  • Code (HumanEval): Sonus-1 Pro (w/ Reasoning) scores 90.0%, a testament to its powerful coding abilities.

  • Code (LiveCodeBench): Sonus-1 Pro (w/ Reasoning) scores 51.9%, displaying solid performance in real-world code environments.

  • Math (GSM-8k): Sonus-1 Pro (w/ Reasoning) reached 97.0% on the challenging math test.

  • Code (Aider-Edit): Sonus-1 Pro (w/ Reasoning) shows impressive performance in code editing by achieving 72.6%.

Notably, Sonus-1 Pro leads the field in various benchmarks and is particularly strong in reasoning and mathematical problems. This shows its ability to outpace other proprietary models.

Capabilities

Sonus-1 demonstrates significant performance capabilities, comparable to the most advanced proprietary models worldwide. See the benchmark results below:


Where To Try Sonus-1?

You can access the Sonus-1 suite of models at chat.sonus.ai.

What's Next?

We're committed to developing high-performance, affordable, reliable, and privacy-focused LLMs. In our commitment to expanding the Sonus-1 series, we're working towards a stable release of additional models that can tackle even harder problems.