Apertus 1.5 8B

Apache 2.0 open model with public weights, training data, and methods. This is the current verified 8B family member; Apertus-8B-2509 is an older version retained separately below.

Score calculation

(7 + 7 + 9) ÷ 3 = 7.7

3 of six dimensions rated. Unrated dimensions and BDS status are excluded.

Full scoring method ↗

Setup details

Apertus model-card claims are developer statements. Its charter/initiative-level mission does not establish downstream use, energy per query, or worker conditions.

Official Apertus documentation lists Apertus 1.5 8B and 70B as current releases, superseding the initial 2025 Apertus 1.0/8B-2509 line. Checked 2026-10-23.

Energy

↑

The Apertus 1.5 technical report says the 8B model received 4T additional pretraining tokens beyond Apertus 1.0 and added multimodal support. The card and report provide no 1.5-8B GPU-hours, device power, kWh, CO2e, water, or standardized inference-energy figure. Earlier Apertus 1.0 70B compute estimates must not be attributed to this 8B checkpoint. Local serving instructions show deployability, not low energy use.

Reasoning & sources

The 1.5 report documents additional training tokens but publishes no 8B energy or emissions total. Energy for an earlier 70B run cannot be transferred to this checkpoint; checkpoint-specific compute and carbon data would be needed for a comparison.

No 1.5-8B training energy/emissions or inference energy measurement; the additional 4T tokens describe training scale, not electricity.

Training rights

↑

The Apertus project describes open training data and methods and says it respects data-owner opt-outs, including retrospective opt-outs. These are developer claims; complete documentation and technical report are available, but no independent rights audit was found.

Reasoning & sources

The project says its dataset is open and supports retrospective data-owner opt-outs, and links to training materials. Those are concrete governance commitments, though they remain developer statements and do not constitute an independent rights audit.

Published corpus and rights-handling disclosures are unusually detailed, but remain developer-reported and lack independent rights verification; score reflects transparency with that material limitation.

Verify whether the 1.5 release preserves the same opt-out mechanisms and dataset coverage as 1.0.

Privacy

↑

Self-hosted open weights allow prompts to remain with the operator. The earlier 8B-2509 card says no PII output filter was provided and recommends applying a periodically updated filter; that caveat is version-specific and cannot be assumed fixed in 1.5.

Reasoning & sources

Apertus 1.5 can be self-hosted, but the documented absence of an output PII filter in the earlier 8B card shows local control does not automatically prevent personal-data disclosure. Operators should add appropriate filtering and evaluate their own logs and access controls.

Public weights enable local deployment and operator control over prompts, but that is a deployment option rather than a guaranteed privacy safeguard; hosted services and downstream configurations require separate review.

Need current 1.5 output-filter status and hosting privacy details.

Openness

↑

Official Apertus docs list 1.5 8B and 70B under Apache 2.0; the project describes architecture, weights, training data, and methods as accessible. Check the specific release's full artifacts before claiming full reproducibility.

Reasoning & sources

The project publishes weights under Apache 2.0 and describes model methods and training data, allowing reuse and inspection. This is strong openness, while an independent audit of data completeness and rights is still absent.

Official project describes Apache 2.0 access to weights, data, and methods.

Confirm complete 1.5 training intermediates and datasets are published.

Worker treatment

↑

public technical materials reviewed identify research teams and training hardware but do not disclose data annotator pay, contract conditions, or worker protections.

Reasoning & sources

Apertus documentation names research teams and infrastructure but does not report data annotator pay or contractor conditions. Research-team attribution is not workforce transparency; ask for details about any human-curated data before drawing a labor conclusion.

No workforce or annotation labour disclosure found.

Apertus documentation and current releases ↗

Government & military

↑

Apertus lists its 1.5 8B checkpoint under Apache 2.0 and links downloadable weights. Independent use is possible without a CSCS or EPFL service contract; the project release pages do not document military use.

Reasoning & sources

Apertus’ official documentation labels the 1.5 8B release Apache 2.0 and links directly to its weights. That means downstream users can deploy it without the research partners as a service provider; it is not evidence of any military deployment.

Needs specific downstream customer/procurement evidence.

BDS status

↑

We did not find Swiss AI Initiative (EPFL, ETH Zurich, CSCS) among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.

Reasoning & sources

Swiss AI Initiative (EPFL, ETH Zurich, CSCS) is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

All sources