Announcing Charlie 1.5: Smart, Small and Secure AI

Redefining AI for Enterprises and Innovators

Today we introduce Charlie 1.5, our most advanced 7B parameter small language model (SLM) designed to meet enterprise needs. This latest release excels in retrieval-augmented generation, extended context processing, and Agentic workflows, while maintaining efficient resource usage.

Charlie 1.5 emphasizes efficiency, accuracy, and adaptability. Designed for flexible deployment—whether on-premises or within private cloud environments—Charlie 1.5 delivers robust performance with minimal infrastructure demands. Its enhanced accuracy, extended context processing, and specialized function-calling features enable improved productivity, creativity, and growth across a wide range of business applications.

Competitive Intelligence

Charlie 1.5 stands as our most advanced model, ranking first or among the top three when compared to its SLM counterparts, models under 10B parameters, across widely recognized model benchmarks. Charlie 1.5 has near human-like comprehension in tackling complex tasks showing strong results on MMLU (assessing undergraduate-level knowledge across multiple domains), GPQA (evaluating graduate-level reasoning capabilities), GSM8K (measuring mathematical problem-solving skills), and other relevant domains.

Lightning Results & Longer Context

Charlie 1.5 delivers near real-time performance capabilities, enabling dynamic applications such as interactive customer support, predictive text completion, and instantaneous data processing. The model operates with an expanded context window of 128,000 tokens while maintaining speeds of up to 200 tokens per second and achieving first-token latency of 0.08 milliseconds. Performance metrics are projected to improve following the initial deployment phase.

Charlie 1.5 achieves 10x the speed of its predecessor while exhibiting enhanced cognitive capabilities across most use cases. The model demonstrates particular proficiency in applications requiring near instantaneous processing, making it ideal for cases in information retrieval and sales process automation. 

Fewer Refusals and Uncensorship

Charlie 1.5 shows improvements in censorship and refusals, offering greater flexibility and nuance in addressing user requests compared to our earlier model. While adhering to minimum safety standards, it demonstrates more measured judgment in handling edge cases, resulting in fewer unnecessary refusals. The model’s enhanced understanding of context and intent allows it to respond effectively to a broader range of appropriate queries that other systems might misinterpret or reject.

Improved Accuracy

To ensure reliable performance at scale, we further evaluated Charlie 1.5 using a comprehensive set of hard, factual questions designed to expose potential vulnerabilities. This testing is crucial, as businesses of all sizes rely on language models for productivity and assistance. Our evaluation categorized responses into three types: correct answers, incorrect answers, and unsure. When tested against these challenging, factual questions, Charlie 1.5 demonstrated remarkable improvements over Charlie 1, achieving twice the accuracy of its predecessor while reducing instances of incorrect responses.

Powering Agentic Workflows

Charlie 1.5 also features enhanced function-calling capabilities that can enable seamless integration with external systems, databases, and APIs. Through specialized training on diverse integration patterns and automation scenarios, the model efficiently handles structured data exchange and complex workflows. This specialty can allow Charlie 1.5 to serve as a robust foundation for enterprise automation and intelligent agent systems, executing tasks with minimal oversight.

On-Prem and Private Deployments

With privacy and security becoming a bigger concern in many industries, businesses are looking more than ever to deploy private models within their infrastructure. Charlie 1.5 is designed for on-premises and private cloud environments, ensuring data privacy with minimal GPU requirements. Its smaller footprint supports customization for industry-specific workflows and brand needs, enabling seamless integration with existing infrastructure. 

In-context learning

We further improved Charlie 1.5 at in-context learning—the ability to adapt its responses based on examples provided within the prompt. This capability manifests in several key approaches: one-shot learning, where a single example demonstrates the desired input-output pattern; few-shot learning, which uses multiple examples to establish more complex patterns; and including chain of thought (CoT) reasoning, where the mode can break down its problem-solving process into explicit steps.

RAG ready

Charlie 1.5 achieves a 60% improvement over Charlie 1 in retrieval augmented generation (RAG), enabling more relevant and contextually informed outputs. This boosts Charlie’s accuracy by integrating external knowledge into response generation, ensuring outputs are grounded in up-to-date and contextually appropriate information.

Elevated Business Use Cases

From engaging blog posts and attention-grabbing social media updates to precise advertisements and customized email campaigns, Charlie 1.5 is optimized to generate high-impact, platform-specific content designed to resonate with target audiences.

This specialization for business use cases is the result of fine-tuning Charlie 1.5 on industry-specific datasets, including marketing materials, brand voice guidelines, and customer interaction examples. Additionally, the model has been trained to analyze audience preferences and adapt tone, style, and messaging to align with specific goals, such as driving engagement or boosting conversions. 

Learn AI in 5 minutes a day

What’s the secret to staying ahead of the curve in the world of AI? Information. Luckily, you can join 800,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.

Sign up to start learning.

The Future of your company starts with Charlie 1.5

We believe Charlie 1.5 is more than an AI model—it’s a catalyst for innovation, efficiency, and growth. We invite you to discover how Charlie 1.5 can transform your business operations and support your strategic objectives. 

For Paying Platform users, you can access Charlie 1.5 via your Charlie API Tool.

If you would like to discuss how Charlie can be adapted to supercharge your business, email [email protected] to discuss!