Koala
Berkeley's open-source chatbot trained on dialogue data from real users — excellent conversational naturalness.
Overview
Koala is a dialogue model created by researchers at UC Berkeley's BAIR (Berkeley Artificial Intelligence Research) lab, fine-tuned from Llama on a curated dataset of high-quality conversational examples including ShareGPT, open-source chatbot datasets, and Wikipedia Q&A. Released in April 2023, it was specifically designed to learn from diverse real human conversation rather than synthetic instruction data.
The key differentiator was the emphasis on conversational quality and naturalness over benchmark performance. Koala was trained to produce contextually appropriate, multi-turn conversation that felt like talking to a knowledgeable human rather than an AI completing tasks. The training data curation focused on quality and diversity of conversation styles.
Koala was evaluated as comparable to Vicuna and approaching ChatGPT on subjective conversational quality metrics, despite being a research model trained at UC Berkeley rather than a commercial product. As with other early 2023 open models, it has been surpassed by more recent releases but remains an important reference point in understanding the evolution of open-source conversational AI.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 8 May 2026 and may be out of date or incomplete.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
Koala 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).
Koala is a research chatbot released by Berkeley's BAIR lab in 2023, fine-tuned on LLaMA using web-sourced dialogue data, and it was an important early demonstration that smaller, cheaply-tuned open models could approach the quality of much larger systems. As an academic milestone in the open-LLM story, it is genuinely significant and worth understanding.
It is not, however, a living product: there is no maintained hosted service, no API, no commercial support, and it has long been superseded by far better open models (Llama 3.x, Mistral, Qwen and others). Treat it as historical reference and a research artefact rather than something to deploy. For an ASEAN team building today, the relevance is educational only.
Notable facts
- Koala's training data included real human conversations from real social media and Q&A platforms, giving it exposure to a wider range of conversational styles than synthetic instruction data alone.
- Berkeley researchers found that Koala beat Alpaca on 44% of human evaluation comparisons while losing on only 26%, demonstrating that conversational data quality matters for chatbot applications.
- The naming of Koala, along with Llama, Alpaca, Vicuna, and others, established the unofficial 'zoo' naming convention for open-source language models.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Koala's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Koala 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 Koala directly →
Spotted something out of date? Suggest an update →
Alternatives to Koala
More in LLMs & Chat
