Backboard.IO Achieves Record-Breaking 90.1% Accuracy on LoCoMo, Setting a New Standard for AI Memory Systems
Backboard scores as highest in every AI memory category.
AI infrastructure company building stateful and portable memory systems, achieves highest score ever for measuring long-term contextual memory in AI models.
Under standardized testing conditions, Backboard reached 90.1% overall accuracy, surpassing all prior public results. The evaluation was performed using the official LoCoMo task suite, GPT-4.1 as the judge model, and identical environmental parameters applied across all participants.
“This result validates our belief that memory should be a foundational layer of AI,” said Rob Imbeault, CEO and founder of Backboard and former co-founder of Assent Compliance, a Canadian SaaS unicorn. “We’re very excited about what we have in our memory product and love seeing how developers are using it.”
Why It Matters
Traditional AI systems are stateless, which means each prompt is treated independently, forcing developers to reconstruct context manually. Backboard’s stateful memory architecture allows AI agents to persist, recall, and reason across thousands of interactions, creating continuity over time without retraining or prompt-stuffing.
Backboard’s API enables developers to integrate persistent, portable memory across 2,200 + LLMs, vector DBs, and embedding models. Memory can move seamlessly between models and sessions, making the system architecture-agnostic and highly configurable.
Technical Highlights
Benchmark: LoCoMo v1.2 (standard task set A)
Judge Model: OpenAI GPT-4.1
Accuracy Score: 90.1 % (highest recorded)
Evaluation Environment: identical parameters, zero fine-tuning bias
Memory Mode Tested: Auto (search + write)
Dataset Transparency: full evaluation logs and methods to be released publicly
“We’re not interested in gaming benchmarks, we want to evolve them,” said Imbeault. “Our goal is memory so good that current benchmarks can’t fully capture its capability.”
Open Verification
Backboard will release its full LoCoMo dataset, evaluation logs, and API endpoint for public verification, inviting the AI research community to reproduce results under the same conditions.
Availability
The Backboard API is currently in alpha with select developers and research partners. Public access and documentation are available at www.backboard.io.
About Backboard
Backboard is an AI infrastructure platform enabling stateful systems with persistent memory and configurable multi-model routing. Its mission is to make AI memory portable, secure, and infinitely scalable across every LLM environment.
About LoCoMo
LoCoMo (Long Context Memory) is an open benchmark that evaluates the ability of AI systems to retain and apply information across extended multi-turn interactions, measuring accuracy, contextual consistency, and reasoning depth.
Maya Ellis
Backboard IO
hello@backboard.io
Visit us on social media:
LinkedIn
YouTube
X
Distribution channels: Business & Economy, IT Industry, Science, Technology, Telecommunications
Legal Disclaimer:
EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.
Submit your press release