smaug-agentic: LiveBench scores and how to run it with BrainOutput
smaug-agentic is a model from Abacus.AI. In the LiveBench release of 2026-06-25 (fetched 2026-10-03) it ranks #13 of 66, with a global score of 79.51.
What is smaug-agentic?
smaug-agentic is listed on LiveBench, a public, continuously refreshed benchmark that scores models on seven categories: reasoning, coding, agentic coding, mathematics, data analysis, language and instruction following. Its strongest category in this snapshot: agentic coding (#7 of 66).
Its weights are published: it can run on hardware you control, with no per-token meter and no vendor who can change the model underneath you.
The model at a glance
| Vendor | Abacus.AI |
|---|---|
| Weights | Open-weight (weights published) |
| Licence | Custom (Kimi K3 licence, inherited) |
| Release date (per LiveBench) | — |
| LiveBench identifier | smaug-agentic |
LiveBench scores, dated
Score and rank out of 66 models, per category, in the LiveBench release of 2026-06-25 fetched 2026-10-03.
LiveBench, release 2026-06-25 · livebench.ai · 66 models · fetched 2026-10-03
Overall rank in this LiveBench snapshot.
Global score: the mean of the seven categories.
Gap to #1, Claude Fable 5.1 Max Effort (83.41).
| Category | Score | Rank |
|---|---|---|
| Reasoning | 90.27 | #13 / 66 |
| Coding | 82.47 | #10 / 66 |
| Agentic coding | 64.65 | #7 / 66 |
| Mathematics | 83.92 | #56 / 66 |
| Data analysis | 79.90 | #9 / 66 |
| Language | 84.36 | #17 / 66 |
| Instruction following | 70.99 | #25 / 66 |
| Global (mean of the 7 categories) | 79.51 | #13 / 66 |
Method note: the global score is the mean of the seven category averages, computed by us from LiveBench's published file; it may differ by a few decimals from the site's own display. Equal scores keep LiveBench's file order. Benchmarks measure benchmark tasks — during an AI diagnostic we re-run the comparison on your own workloads.
Running smaug-agentic with BrainOutput
BrainOutput routes each task to a model by confidentiality: sensitive work stays on a local or private model, and everything else goes to a cloud model only when your policy allows it.
Locally, on your hardware
Available
The weights run on a machine on your premises. The right hardware depends on model size and concurrent users: we size it during the AI diagnostic.
In a private cloud
Available
The same model, served from dedicated private inference, under the same routing policy as on-premises.
With your own key (BYOK)
Available
If a hosted service you already pay for offers this model, plug in your key: BrainOutput sends it only non-sensitive work, and only when your policy allows it.
Frequently asked questions
Is smaug-agentic an open-weight model?
Yes: its weights are published. You can run it on your own hardware or in a private cloud; licence terms vary by model and we check them before any deployment.
Can we use smaug-agentic with BrainOutput?
Yes: locally on your hardware, in a private cloud, or through your own API key (BYOK). Hardware sizing is done during the AI diagnostic, for your workload — we do not publish figures we have not measured.
Where do these scores come from?
From LiveBench (livebench.ai), release 2026-06-25, file fetched 2026-10-03. The global score is the mean of the seven categories, computed by us. The ranking changes monthly; we re-verify at diagnostic time.
Find out whether smaug-agentic fits your work
An AI diagnostic maps your processes, classes what is sensitive, and benchmarks current models against your real tasks — dated numbers, yours this time.
Sources
- LiveBench — leaderboard, release 2026-06-25 — accessed 2026-10-03
- smaug-agentic — vendor and licence — accessed 2026-10-04