Apertus
Switzerland's fully open LLM: weights, data and training recipes all published
Overview
Apertus is a fully open large language model from the Swiss AI Initiative (EPFL, ETH Zurich and CSCS), released in September 2025 in 8B and 70B sizes under Apache 2.0, with its training data and recipes published. Apertus 1.5 (July 2026) adds image understanding, a thinking mode and a 262,144-token context window, and Apertus Mini adds small distilled models down to 0.5B.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 19 Oct 2026 and may be out of date or incomplete.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.
What you can produce with Apertus
- Apache 2.0 weights in 8B and 70B, with training data, code and recipes published
- Image understanding, an optional thinking mode and a 262,144-token context (Apertus 1.5)
- Small distilled models from 0.5B to 4B, with MLX and vLLM builds (Apertus Mini)
- Text across more than 1,000 languages, with about 40% of training data non-English
- Per-token hosted access from Swiss providers including Swisscom and Infomaniak
- EU AI Act transparency documentation and a published model charter
Apertus is the model to pick when provenance matters more than leaderboard rank: weights, data and recipes are public, web opt-outs were honoured retroactively, and Apache 2.0 allows commercial use. Version 1.5 closed real gaps with vision, a thinking mode and long context, and Swiss providers now host it per token.
Its own team says it trails frontier open-weight models, and Southeast Asian languages get no special attention. Choose it for auditable, sovereign deployments in Europe, or as a reference for open training practice.
What we found
Apertus is the open language model of the Swiss AI Initiative, built by EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) and released on 2 September 2025 in 8B and 70B sizes. Its name is Latin for "open", and the claim is meant literally: weights, training data, code and recipes are published, and the models are under Apache 2.0, which allows commercial use. Training used only publicly available, permissively licensed data, honoured robots.txt opt-outs retroactively, filtered personal data and used a "Goldfish" objective to suppress verbatim memorisation. The project says it covered more than 1,000 languages; its ACL 2026 paper counts over 1,800, with about 40% of the 15 trillion tokens non-English, including Swiss German and Romansh.
The family has grown quickly. Apertus Mini (version 1.1, June 2026) added sixteen distilled and quantized models from half a billion to four billion parameters, with MLX builds for Apple devices. Apertus 1.5, whose weights arrived on 24 July 2026, continued training the 8B on four trillion more tokens and the 70B on two trillion, adding image understanding, experimental audio input, an optional thinking mode and a 262,144-token context window. The 1.5 checkpoints ask users to accept terms on Hugging Face, unlike the 1.0 release. The team plans a much larger Apertus 2.0 for the first quarter of 2027.
The team states performance modestly: Apertus 1.5 still trails frontier open-weight models, though it says the 8B leads similarly sized open models on instruction-following and multilingual tasks, and its multimodal scores sit near Gemma 4 12B. The project reports more than four million downloads across its releases.
You need not host it yourself. Swisscom, Infomaniak, Phoeniqs and Safe Swiss Cloud sell per-token access from Swiss data centres, while Public AI and Featherless serve it internationally; Swiss agency Liip notes that Public AI's chat lets anyone try the 8B without signing in. The Canton of Ticino is trialling an in-house translation tool built on a fine-tuned 8B.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Apertus's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Apertus 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 Apertus directly →
Spotted something out of date? Suggest an update →
Alternatives to Apertus
More in LLMs & Chat
