Logo AnoveLogo AnoveExplAIn
Pionniers de l'IA
Retour au répertoire

baidu/Qianfan-OCR

par Baidu

AI Model
Faible confiance

Baidu/Qianfan-OCR is a 4B-parameter end-to-end multimodal model for document intelligence, unifying OCR, layout analysis, and document understanding in a single vision-language architecture. It directly converts images to Markdown and supports tasks like table extraction, chart understanding, key information extraction (KIE), and multilingual OCR (192 languages). Powered by Qianfan-ViT (vision encoder) + Qwen3-4B (language model), it achieves #1 rankings on OmniDocBench v1.5 (93.12), OlmOCR Bench (79.8), and KIE benchmarks (87.9). The model introduces Layout-as-Thought, an optional thinking phase (⟨think⟩ tokens) for structured layout recovery, and delivers high throughput (1.024 PPS on A100 with W8A8 quantization). It is open-source (Apache 2.0 License) and deployable via transformers or vLLM.

Visiter le site web

Score de transparence

Pondéré sur quatre piliers · mis à jour le June 30, 2026

Note globale
32% pondéré
ABCDF
Modèle
—
Pas encore évalué
Infrastructure
—
Pas encore évalué

Capacités

Image To Text

Informations sur le fournisseur

Informations complètes sur le fournisseur/prestataire de cette application IA

BaiduVoir tous les produits →
Baidu, Inc.(Exerçant sous le nom : Baidu)
Informations de contact
www.baidu.com/
intlcomm@baidu.com
(+86 10)5992 8888
Adresse enregistrée
Baidu Campus, No. 10 Shangdi 10th Street,, Beijing, Haidian District, 10085, CN

Informations fournisseur (Règlement IA de l’UE)

Statut de vérification:
Verified
(Dernière vérification : August 5, 2026)
Baidu, Inc.(Exerçant sous le nom : Baidu)
Baidu Campus, No. 10 Shangdi 10th Street,, Beijing, Haidian District, 10085, CN
intlcomm@baidu.com(+86 10)5992 8888
Documents de conformité
Marquage CE: Not Applicable
No product-safety style post-market monitoring programme in the EU AI Act sense has been identified for this vendor. The available monitoring surfaces are: periodic financial disclosure to the SEC (Form 20-F, Form 6-K) and to HKEX; the investor relations newsroom at https://ir.baidu.com/press-releases; and Chinese domestic regulatory processes, including the Cyberspace Administration of China approval regime for publicly available generative AI services, under which Baidu's ERNIE Bot was among the first batch of approved services in August 2023. No published incident-reporting channel, model-update changelog, vulnerability disclosure policy or serious-incident process specific to the AI products was located. Absence of located evidence is not evidence of absence; this field should be revisited if any Baidu product enters scope as a deployed system.
Documented and structural limitations relevant to third-party risk assessment: (1) Jurisdiction and transfer. Operations and hosting are in mainland China under PIPL, the Data Security Law and the Cybersecurity Law; no EU data residency option has been identified for the consumer AI products, so any EU personal data would require a GDPR Chapter V transfer assessment. (2) No EU establishment and no EU AI Act authorized representative identified, which is material if any Baidu AI system falls within EU AI Act scope. (3) No enterprise DPA, standard contractual clauses, or SOC 2 / ISO 27001 certification located for the consumer AI product lines. (4) Model opacity: no model cards, training-data documentation or independent evaluation results published for the generative features. (5) IP provenance: generative products grounded in user-uploaded third-party content, notably Baidu Wenku, carry unresolved copyright exposure, and the Baidu Wenku upload model has been the subject of copyright litigation in China. (6) Language: interfaces and most documentation are Simplified Chinese only, which is an operational and reviewability constraint for a Netherlands-based deployer. (7) Product volatility: the AI line changes rapidly (multiple GenFlow major versions between April 2025 and Q1 2026, a Kunlunxin spin-off and separate listing in progress), so records against this vendor go stale quickly.

Risques potentiels

1 points d'attention identifiés

1

Révision recommandée avant utilisation

Ces points d'attention sont automatiquement identifiés sur la base d'informations publiquement disponibles sur le fournisseur et les données du catalogue IA. Les risques réels peuvent varier selon votre cas d'usage et votre implémentation spécifiques.

Conformité et risques

Obtenez des informations sur les risques en exécutant des évaluations sur cette application IA.

Vous travaillez chez Baidu ? Revendiquez cette fiche pour corriger ou compléter les données.

Ajouté : June 30, 2026
Mis à jour : June 30, 2026

Prêt à gérer vos applications IA ?

Suivez, évaluez et gouvernez vos applications IA avec Anove.

Logo AnoveLogo AnoveAnove International B.V.

Nous aidons les organisations à découvrir et gérer les applications IA de manière responsable. Votre source de confiance pour la gouvernance et la conformité IA.

Liens rapides

  • Parcourir le répertoire
  • FAQ
  • Se connecter

Ressources

  • Documentation
  • Statut
  • Anove.ai

Conditions

  • Politique de confidentialité
  • Conditions générales
  • Divulgation responsable
  • Mention légale IA
  • Politique de sécurité

© 2026 Anove International B.V. Tous droits réservés.