오픈소스 LLM.
열린 코드, 나의 선택.
오픈소스 대규모 언어 모델은 공개된 코드로 만들어진 AI 시스템입니다. 누구나 작동 방식을 검토하고 감사하며 확인할 수 있습니다. 아누마는 기본적으로 프롬프트로 자체 모델을 학습하지 않으며, 데이터 보관은 해당 제공업체의 약관을 따릅니다.
How open-source LLMs work
A large language model (LLM) is the AI engine behind tools like ChatGPT and Kimi. It processes your text and generates responses.
An open-source LLM is one where the source code and model weights are publicly available. This means:
- Anyone can inspect the code to verify what it does with your data
- Independent researchers can audit it for safety and bias
- No single company controls how it works
- It can run on your own hardware or on infrastructure you trust
Examples on Anuma: Kimi (Moonshot AI), MiniMax (MiniMax), Cerebras Qwen 3 (Cerebras), GLM (Zhipu).
Open-source vs closed-source LLMs
Two approaches to AI, both included in every Anuma plan.
Open-Source LLMs | Closed-Source LLMs | |
|---|---|---|
| Source code | Publicly available, auditable | Proprietary, not inspectable |
| Who builds them | Open research labs, communities | OpenAI, Anthropic, Google, xAI |
| Data retention | Third-party providers run them. Their own terms govern retention | Per provider terms. Encrypted in transit and at rest |
| Where they run | Your device or trusted infrastructure | Provider servers with enterprise hardware |
| Transparency | Full. You can verify everything | Limited. You trust their privacy policy |
| Best for | Coding, math, privacy-sensitive work | Creative writing, complex reasoning, nuanced analysis |
| Quality | Rivals closed-source on coding/math | Generally leads on creative/nuanced tasks |
| On Anuma | All included in every plan | All included in every plan |
Why open-source LLMs matter
Transparency, independence, and privacy by design.
Transparency
You don't have to trust a company's privacy policy. You can read the code yourself. Open-source means model behavior you can inspect, with no black box.
Independence
Open-source LLMs aren't controlled by any single company. If one provider changes their terms, raises prices, or adds restrictions, open-source alternatives are always available.
Privacy by design
On Anuma, open-source LLMs are run by third-party providers whose own terms govern retention and training. Anuma does not use your prompts to train its own models by default, and conversations are encrypted in transit and at rest.
Open-source LLMs on Anuma
Access the strongest open-source LLMs from a single account. New models are added as they launch.
Kimi
Open-SourceEnhanced reasoning and coding. Among the strongest open-source options for analytical tasks, complex problem-solving, and technical writing.
MiniMax
Open-SourceRivals frontier closed-source LLMs on key benchmarks. The most capable general-purpose open-source LLM, strong across conversation, analysis, and content generation.
Cerebras Qwen 3
Open-SourceA massive 235 billion parameter model with ultra-fast inference. One of the largest open-source LLMs available.
GLM
ZhipuStrong multilingual performance and general reasoning. Particular strength in non-English languages.
Who uses open-source LLMs?
Open-source LLMs serve anyone who values transparency, privacy, and control.
Developers
Open-source LLMs like Kimi handle coding and technical work well. Debug, review, and generate code with models you can inspect.
Privacy-conscious professionals
Lawyers, doctors, financial advisors handling sensitive client data. Open-source models you can inspect, and Anuma does not use your prompts to train its own models by default.
Researchers
Run analysis with models you can inspect. Anuma does not use your prompts to train its own models by default. Full transparency into model behavior.
Businesses with sensitive data
Product roadmaps, financial models, and competitive analysis, handled with models whose code you can inspect.
Anyone who values knowing how their tools work
If you want certainty about where your data goes, open-source gives you that.
Private by design, with Anuma model training off by default.
Anuma does not use your prompts and responses to train its own models by default. Open-source LLMs are run by third-party providers whose own terms govern retention and training, and Anuma does not control that.
Your memory still works.
Open-source LLMs on Anuma access your Memory Vault just like closed-source LLMs. Your context carries across. Privacy doesn't mean starting over.
Switch to closed-source LLMs anytime.
Need ChatGPT or Claude for a specific task? One tap. Your memory follows. Use open-source for sensitive work, closed-source for capability-heavy tasks.
Kimi
Claude
ChatGPTFrequently asked questions
Everything you need to know about open-source LLMs on Anuma.
A large language model where the source code and weights are publicly available. Anyone can inspect, audit, and run the model independently. This is different from closed-source LLMs like ChatGPT or Claude where the code is proprietary.
Kimi, MiniMax, Cerebras Qwen 3, GLM. All are included on every plan, including the free tier.
Open-source LLMs match or beat closed-source on coding and math tasks. Closed-source LLMs generally lead on creative writing and nuanced reasoning. The gap is narrowing rapidly.
Private by design, with limits. Open-source LLMs are run by third-party providers whose own terms govern retention and training. Anuma does not use your prompts to train its own models by default, and conversations are encrypted in transit and at rest.
Not necessarily. MiniMax competes with frontier closed-source LLMs on general tasks. For creative writing and nuanced reasoning, closed-source LLMs still have an edge, but the gap is closing.
Yes. Switch in one tap. Your Memory Vault carries your context across both. Use open-source for privacy-sensitive work and closed-source when you need maximum capability.
Open-source LLMs are included on every Anuma plan, including the free tier. You don't need a separate subscription.