Anthropic has still not open-sourced any LLM as of 2026

As of 2026, Anthropic, one of the leading players in the artificial intelligence sector, has never open-sourced its large language model (LLM). This stance is an extension of its consistent strategy since its founding in 2019: favoring a closed-system approach to preserve model performance, minimize security risks, and sustain its commercial advantage.

Anthropic’s Closed-System Policy: Why?

Anthropic holds a significant position in the market with its Claude series of models. However, unlike developments such as OpenAI partially open-sourcing its GPT series, Meta fully open-sourcing its Llama series, and Google offering partial access to its Gemini models, Anthropic has always built its ecosystem on a closed-model framework. Dario Amodei, one of the company’s founders, stated during a 2024 panel: “Open-sourcing, especially for powerful models, can increase risks of malicious use and arms races.” This perspective remains valid as of 2026.

Comparisons Within the Industry

As of 2026, the trend toward open-sourcing large language models is rapidly accelerating. Meta’s Llama 3, Llama 3.1, and Llama 3.2 versions have gained popularity among academic and commercial users, while Mistral AI and Alibaba’s Qwen series also hold substantial shares in the open-source model landscape. Anthropic, however, continues to resist this trend. Its Claude 3.5 Sonnet, Claude 3.7 Sonnet, and Claude 3.7 Opus models are accessible only via API and can be used exclusively through Anthropic’s own infrastructure and partners.

Learning and Security: Arguments Against Open-Sourcing

Anthropic summarizes its reasons for not open-sourcing into three key points:

• Security: Releasing full model versions increases risks such as fraud, manipulation, and malicious content generation.
• Commercial Sustainability: Open-sourcing models undermines the revenue streams of the company’s service-based business model.
• Control and Improvement: A closed system enables continuous optimization of the model for safety and ethical alignment.

This approach is closely monitored by regulatory bodies in Europe and the United States. In 2025, the EU introduced transparency requirements for large language models under the AI Act; however, Anthropic argues that its API-delivered models qualify as “in-system” and therefore fall outside the scope of these regulations.

The Future: Pressure for Open-Sourcing Is Growing

As of 2026, academic circles and the open science community are increasing pressure on Anthropic’s open-source policy. Researchers from institutions such as MIT, Stanford, and Berkeley have stated they require full access to the training data and outputs of Anthropic’s models for analysis. Anthropic, however, has so far preferred to emphasize “ethical transparency” over “model transparency,” publishing “Constitutional AI” reports that explain the model’s decision-making mechanisms instead.

Anthropic’s stance highlights how fragile the balance between closed and open systems remains in the AI world. In the coming years, pressures from regulators and demands from the open-source community may force the company to reevaluate its strategy. Yet as of 2026, Anthropic’s commitment to not open-sourcing remains one of the most prominent and consistent policies in the industry.