Smaug by Abacus
Abacus.AI's 2024 open-weight fine-tunes, now legacy
Overview
Smaug is Abacus.AI's 2024-era family of open-weight fine-tunes (Qwen- and Llama-based) that briefly topped the Hugging Face Open LLM Leaderboard. No releases since mid-2024; the weights remain freely available on Hugging Face.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
ASEAN Perspective
Smaug by Abacus in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Note: Smaug's original hosted product page no longer resolves; only the open-weight models remain available.
Smaug is a 2024 artifact. Abacus.AI's fine-tune family — Smaug-72B was the first open-weight model to average over 80 on the old Hugging Face Open LLM Leaderboard in February 2024, with a Llama-3-70B variant that May — proved that clever preference optimisation (their DPOP technique) could squeeze real gains from strong open bases. But there have been no new Smaug releases since, the abacus.ai/smaug page now 404s, and Abacus's energy has moved to ChatLLM, DeepAgent, and newer fine-tunes. Current open weights from Llama, Qwen, and DeepSeek outclass it across the board. It's only worth a look for researchers studying fine-tuning methods or reproducing 2024 baselines; the weights remain free on Hugging Face for anyone with the GPUs to serve them.
What people say
For a few months in early 2024, Smaug-72B was the best open-weight model in the world — the first to average over 80 on the Hugging Face Open LLM Leaderboard, built by fine-tuning Qwen-72B with Abacus.AI's DPO-Positive (DPOP) technique. A Smaug-Llama-3-70B-Instruct followed in May 2024 with conversational scores nipping at GPT-4 Turbo's heels. As a demonstration that a small applied-AI shop could out-tune the giants, it worked.
As a 2026 directory listing, it's a museum piece. There have been no new Smaug releases since mid-2024; Abacus.AI's public energy went to Dracarys, its coding fine-tunes, and then to products — ChatLLM Teams and DeepAgent — that package frontier models rather than tune open ones. The old abacus.ai/smaug page now returns a 404. The weights live on via Hugging Face, and the abacusai/smaug GitHub repo mostly hosts the DPOP paper artifacts.
Anyone shopping for open weights today has strictly better options: newer Llama, Qwen, and DeepSeek releases post scores Smaug can't touch, on the modern benchmark suites that replaced the board Smaug topped — Hugging Face retired the original leaderboard partly because benchmark-tuned fine-tunes were gaming it, a debate Smaug itself featured in. The honest use cases left are research into preference-optimisation methods and reproducing 2024 baselines. The weights are free, licence terms follow the underlying base models, and it will run fine if you have the GPUs. Just don't mistake it for a current recommendation.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Smaug by Abacus's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Smaug by Abacus unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Smaug by Abacus directly →
Spotted something out of date? Suggest an update →
Smaug by Abacus in the news
Alternatives to Smaug by Abacus
More in LLMs & Chat
