Vicuna
Fine-tuned open-source chatbot trained on ShareGPT conversations — 90% quality of ChatGPT at minimal cost.
Overview
Vicuna is an open-source chatbot model created by researchers at UC Berkeley, CMU, Stanford, and UC San Diego by fine-tuning Meta's Llama model on conversation data from ShareGPT (user-shared ChatGPT conversations). Released in March 2023, it was among the first open-source models to demonstrate near-ChatGPT quality, evaluated by GPT-4 at approximately 90% of ChatGPT performance.
The training cost for Vicuna was approximately $300 in cloud computing, demonstrating that high-quality instruction-following models could be produced at a fraction of the cost of large commercial model training. This accessibility sparked the 'Local LLM' movement where developers began experimenting with running capable models on their own hardware.
Vicuna 13B and 33B variants were released, with the 13B model running on consumer hardware. The model demonstrated that the combination of a good base model (Llama) and good conversational fine-tuning data (ShareGPT) could produce a surprisingly capable chatbot. While surpassed by later models, Vicuna was historically important as proof that open-source fine-tuning could approach commercial model quality.
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 These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete.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
Vicuna 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).
Vicuna is a historically important open-source chat model released in 2023 by LMSYS, fine-tuned from Meta's LLaMA on user-shared ChatGPT conversations and notable for showing how cheaply a strong open chatbot could be approximated. It was a landmark in the early open-LLM wave and remains a useful reference point and a fixture in the LMSYS evaluation lineage.
It suits researchers and students studying the evolution of open models, not teams looking to deploy in production today. The honest caveat is age: by 2026 Vicuna is comprehensively outclassed by modern open models (Llama 3.x, Qwen, Mistral, DeepSeek and others) on every practical axis, and its non-commercial LLaMA heritage limits real use. There is no managed API; it is weights and a research artifact. Treat it as history, not a current tool.
What people say
$300. That was the reported compute bill for fine-tuning Vicuna-13B in March 2023, and for a while it bought the most talked-about number in AI: '90% of ChatGPT quality,' as judged by GPT-4. The asterisk was doing heavy lifting — GPT-4-as-judge was a brand-new, unvalidated methodology the authors themselves described as a fun, non-scientific evaluation — but the point landed. A student team spanning Berkeley, CMU, Stanford and UCSD had fine-tuned Meta's LLaMA on roughly 70,000 user-shared ChatGPT conversations from ShareGPT and gotten something startlingly usable.
Vicuna's real legacy is institutional. To evaluate it, the LMSYS team needed pairwise human comparisons at scale, so they built Chatbot Arena — which grew into lmarena.ai, the closest thing the industry has to a public leaderboard of record. FastChat, the serving and training code released alongside Vicuna, became widely used infrastructure in its own right. The model line got a friendlier licensing footing with v1.5 on Llama 2 in August 2023, then quietly stopped.
As a tool today, there is no case. The original had a 2K context window, the ShareGPT training data sat in terms-of-service grey territory, and every practical axis — reasoning, instruction-following, languages, context — has been passed many times over by Llama 3.x, Qwen, Mistral and DeepSeek. There's no hosted API; it's weights on Hugging Face and a blog post.
The 3.2 score is honest: this is a museum piece with an outsized wing named after it. Study it for the history; deploy literally anything newer.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- Vicuna was trained in one day for approximately $300 — 100,000x cheaper than the estimated cost of training the original ChatGPT.
- The name 'Vicuna' continues the South American camelid theme started by Meta's Llama model.
- Vicuna's release sparked the LMSYS Chatbot Arena, a crowd-sourced human evaluation benchmark where users compare AI models anonymously.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Vicuna's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Vicuna 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 Vicuna directly →
Spotted something out of date? Suggest an update →
Vicuna in the news
More in LLMs & Chat
