Abacus.AI has released Smaug, a family of three open-weight large language models tuned for long-running enterprise AI agents, the company said on 10 September 2026.
Abacus.AI describes Smaug as a fine-tuning technique that can be applied to any open-source base model. It says the technique improves performance in long-running agentic loops by 15–20% without raising cost, and that open-source models typically cost 10–100 times less to run than frontier models from Anthropic and OpenAI.
Smaug Agentic, based on Kimi K3 and described as a large 2T model, targets complex coding loops. Smaug Flash, fine-tuned on DeepSeek Flash, is built for personal agents that connect to WhatsApp, Telegram and Slack. Smaug Mini is a smaller 27B model for multimodal use cases and lighter reasoning workloads.
All three models are open-weight and available on Hugging Face. Enterprises can host them inside their own cloud VPC, and Smaug Agentic can run on an in-house GPU cluster.
“Open-weight models are rapidly closing the gap to frontier closed models, but still underperform in long-running agent loops,” said Bindu Reddy, CEO of Abacus.AI.
San Francisco-based Abacus.AI is backed by Coatue, Tiger Global and Index Ventures, and says its platform is used by more than 3 million professionals, small businesses and enterprises.