Smaug Agentic
Kimi K3 fine-tune optimized for long-running enterprise agentic loops.
About
Not a new foundation model but a fine-tune: Abacus.AI took Kimi K3 (2.8T total, 104B active) and tuned it for long-running agentic loops, claiming 15–20% better agentic-loop performance at the same inference cost. It shipped September 10, 2026 with open weights on Hugging Face, plus RouteLLM API and VPC self-hosting options.
Abacus.AI reports 94.1 on GPQA Diamond versus Kimi K3's 93.5 — company testing, so treat it as a claim — and it inherits the Kimi K3 License's large-MaaS-business terms.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- gpqa-diamond: 94.1 (Abacus.AI-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Fine-tune of Kimi K3 by Abacus.AI, not a new foundation model; released Sept 10, 2026; Claimed 15-20% better long-running agentic-loop performance at the same inference cost — company claim; GPQA Diamond 94.1 vs Kimi K3 93.5 is Abacus.AI-reported; benchmarks self-tested. License: Inherits the custom Kimi K3 License (large MaaS businesses need a separate agreement); verify terms before commercial use..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 5.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Fine-tune of Kimi K3 by Abacus.AI, not a new foundation model; released Sept 10, 2026
- · Claimed 15-20% better long-running agentic-loop performance at the same inference cost — company claim
- · GPQA Diamond 94.1 vs Kimi K3 93.5 is Abacus.AI-reported; benchmarks self-tested