Saltar al contenido principal

Catálogo de Modelos LLM como Servicio

Visión general​

Cloud Temple LLMaaS ofrece 46 modelos de lenguaje grande cuidadosamente seleccionados y optimizados para satisfacer los requisitos más estrictos de SecNumCloud. Nuestro catálogo cubre todo el espectro, desde micromodelos altamente eficientes hasta modelos extremadamente grandes.

Estadísticas Generales​

MétricaValor
Número total de modelos46 modelos
Contexto mínimo2.048 tokens
Contexto máximo262.144 tokens
ConformidadSecNumCloud ✅ HDS ✅ Soberanía ✅ C5 ✅
Localización100% Francia 🇫🇷

Tarifación​

Tipo de usoPrecio
Tokens de entrada1,90 € / millón de tokens
Tokens de salida8 € / millón de tokens
Razonamiento avanzado8 € / millón de tokens

Modelos de Gran Tamaño​

cogito:32b​

Deep Cogito • 32B parameters • Context: 32,000 tokens

Advanced version of the Cogito model offering significantly enhanced reasoning and analytical capabilities, designed for the most demanding AI analytical applications.

Technical specifications:

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Reasoning Understanding Analysis

Use cases:

  • Multi-factorial scenario analysis with probabilistic evaluation of outcomes
  • Scientific problem solving with formal demonstration of steps
  • High-criticality applications requiring precision and verifiability of results
  • Expert systems in specialized domains (legal, medical, technical)
  • Analysis with multi-step reasoning and full explainability of conclusions

gemma3:27b​

Google • 27B parameters • Context: 120,000 tokens

Revolutionary model from Google offering an optimal balance between power and efficiency, with an exceptional performance-to-cost ratio for demanding professional applications.

Technical specifications:

  • Speed: 21 tokens/second
  • Energy consumption: 6.35 kWh per million tokens
  • License: Google Gemma Terms of Use
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ❌ Security

Tags: Vision Agent Large context

Use cases:

  • Document analysis with extended context up to 120K tokens (approximately 400 pages)
  • Semantic indexing and search in large document databases
  • Simultaneous processing of images and text thanks to multimodal capabilities
  • Structured data extraction from PDFs and scanned documents
  • Integration with external tools via function calling API

glm-4.7-flash:30b​

Zhipu AI • 30B parámetros • Contexto: 120.000 tokens

Versión Flash del modelo GLM-4.7, optimizada para velocidad y eficiencia.

Especificaciones técnicas:

  • Velocidad: 103 tokens por segundo
  • Consumo: 1,41 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ❌ Visión • ✅ Razonamiento • ❌ Seguridad

Etiquetas: Agente Rápido Gran contexto Multilingüe

Casos de uso:

  • Asistentes conversacionales rápidos
  • Análisis de documentos largos (hasta 200k)
  • Tareas de razonamiento con baja latencia

glm-4.7:358b​

⚠️ WARNING: This model is deprecated. Model removed from the catalog on 03/30/2026.

Zhipu AI • 358B parameters • Context: 120,000 tokens

High-performance versatile model developed by Zhipu AI, excelling in logical reasoning, multilingual understanding, and complex tasks.

Technical specifications:

  • Speed : 18 tokens/second
  • Consumption : 7.41 kWh per million tokens
  • License : Apache 2.0
  • Location : FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Reasoning Large Context Multilingual

Use cases:

  • Complex reasoning tasks
  • Long document analysis
  • Advanced conversational assistants

gpt-oss:120b​

OpenAI • 120B parameters • Context: 120,000 tokens

State-of-the-art open-weight language model from OpenAI, delivering strong performance with a flexible Apache 2.0 license.

Technical specifications:

  • Speed: 104 tokens/second
  • Energy consumption: 2.19 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: MoE Agent Reasoning Open-Source Very Large

Use cases:

  • Advanced conversational agents with complex reasoning and tool integration.
  • Applications requiring full transparency in the reasoning process (chain-of-thought).
  • Commercial scenarios needing a permissive license (Apache 2.0).
  • Fine-tuning for specialized tasks requiring a powerful base model.

llama3.3:70b​

Meta • 70B parameters • Context: 132,000 tokens

State-of-the-art multilingual model developed by Meta, designed to excel in natural dialogue, complex reasoning, and nuanced instruction understanding.

Technical specifications:

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Agent Dialogue Multilingual

Use cases:

  • Multilingual chatbots supporting 8 languages simultaneously
  • Execution of complex, chained instructions (prompt chaining)
  • Processing of 60K-token dialogue windows for conversational history
  • Analysis of large legal or technical documents (>100 pages)
  • Generation of structured text with strict adherence to stylistic guidelines

ministral-3:14b​

Mistral AI • 14B parameters • Context: 250,000 tokens

The most powerful model in the Ministral family, designed for complex tasks on local infrastructure.

Technical specifications:

  • Speed: 31 tokens/second
  • Consumption: 4.30 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: High Performance Edge Reasoning Code

Use cases:

  • Solving complex problems locally
  • Coding and engineering assistants
  • Deep document analysis with reasoning

nemotron-3-nano:30b​

NVIDIA • 30B parameters • Context: 250,000 tokens

NVIDIA-optimized model for complex reasoning and tool utilization, deployed with an extended context.

Technical specifications:

  • Speed: 89 tokens/second
  • Consumption: 1.62 kWh per million tokens
  • License: NVIDIA Community License
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Reasoning Large Context

Use cases:

  • Complex autonomous agents with multiple tool calls
  • Logical reasoning and problem solving
  • Long document analysis with precise extraction

olmo-3:32b​

AllenAI • 32B parámetros • Contexto: 65.536 tokens

El primer modelo de razonamiento completamente abierto a esta escala, rivalizando con los mejores modelos propietarios.

Especificaciones técnicas:

  • Velocidad: 19 tokens/segundo
  • Consumo: 7,02 kWh/millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ❌ Herramientas/Agente • ❌ Visión • ✅ Razonamiento • ❌ Seguridad

Etiquetas: Open-Source Gran Contexto Razonamiento Transparencia Código Alto Rendimiento

Casos de uso:

  • Razonamiento complejo y resolución de problemas multi-etapa
  • Desarrollo de software avanzado y generación de código
  • Análisis profundo que requiere transparencia sobre el proceso de toma de decisiones

olmo-3:7b​

AllenAI • 7B parameters • Context: 65,536 tokens

Reference "Fully Open" model, offering complete transparency (data, code, weights) and remarkable efficiency.

Technical specifications:

  • Speed: 37 tokens/second
  • Consumption: 1.65 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Open-Source Large Context Transparent Efficient Maths Code

Use cases:

  • Academic and scientific research requiring full reproducibility
  • Programming tasks and mathematical problem solving
  • Analysis of medium-sized documents with full traceability

qwen-coder-next:80b​

Qwen Team • 80B parameters • Context: 250,000 tokens

State-of-the-art MoE model optimized for code and complex reasoning.

Technical specifications:

  • Speed: 98 tokens/second
  • Consumption: 1.47 kWh/million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Programming MoE Large Context AWQ

Use cases:

  • Advanced programming assistant (repo-scale)
  • Analysis and refactoring of complex code
  • Autonomous software engineering agents

qwen3-2507:235b​

Qwen Team • 235B parameters • Context: 130,000 tokens

Massive MoE model with 235 billion parameters, activating only 22 billion at a time, delivering state-of-the-art performance.

Technical specifications:

  • Speed: 58 tokens/second
  • Energy consumption: 3.93 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: MoE Agent Reasoning Very Large

Use cases:

  • Solving complex mathematical and logical problems
  • Tasks requiring extensive knowledge base
  • Advanced coding assistant
  • In-depth document analysis

qwen3-2507:30b-a3b​

Qwen Team • 30B parameters • Context: 250,000 tokens

Improved version of the non-thinking mode of Qwen3-30B, featuring enhanced general capabilities, broader knowledge coverage, and better user alignment.

Technical specifications:

  • Speed: 104 tokens/second
  • Energy consumption: 1.39 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Agent Large Context MoE Multilingual

Use cases:

  • Complex tasks requiring precise instruction following and logical reasoning.
  • Multilingual applications with extensive knowledge coverage.
  • High-quality text generation for open-ended and subjective tasks.
  • Analysis of very large documents thanks to the 250k-token context.

qwen3-coder:30b​

⚠️ WARNING: This model is deprecated. Recommendation to migrate to qwen-coder-next:80b.

Qwen Team • 30B parameters • Context: 250,000 tokens

MoE-optimized model for software engineering tasks, featuring a very long context.

Technical Specifications:

  • Speed: 104 tokens/second
  • Energy Consumption: 1.39 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Programming Large Context MoE

Use Cases:

  • Software engineering agents for exploring and modifying codebases
  • Generation of complex code with repository-scale understanding
  • Reasoning tasks over extended contexts
  • Code improvement via reinforcement learning

qwen3-next:80b​

Qwen Team • 80B parameters • Context: 250,000 tokens

Next 80B model from Qwen, optimized for large contexts and reasoning.

Technical specifications:

  • Speed : 98 tokens/second
  • Consumption : 1.47 kWh/million tokens
  • License : Apache 2.0
  • Location : FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Agent Reasoning Large Context MoE

Use cases:

  • Advanced conversational agents with tool integration
  • Analysis of very large documents (up to 260k tokens)
  • Code generation and complex tasks requiring structured reasoning

qwen3-omni:30b​

Qwen Team • 30B parameters • Context: 32,768 tokens

Qwen3-Omni 30B is a native multimodal model, capable of understanding text, images, video, and audio within a single stream.

Technical specifications:

  • Speed: 86 tokens/second
  • Consumption: 2.65 kWh/million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ✅ Reasoning • ❌ Security

Tags: Omni Audio Vision Agent Multimodal BF16

Use cases:

  • Fluid multimodal interactions (speaks, sees, listens)
  • Combined video and audio analysis
  • Next-generation intelligent assistants

qwen3-vl:235b​

Qwen Team • 235B parameters • Context: 200,000 tokens

The most powerful multimodal model in the catalog, combining state-of-the-art visual understanding with exceptional reasoning capabilities.

Technical specifications:

  • Speed: 31 tokens/second
  • Consumption: 7.35 kWh/million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Reasoning Large Context NVFP4 Blackwell Vision

Use cases:

  • Automation of complex document workflows (multilingual OCR, structured extraction)
  • Intelligent visual agents for software interaction and GUI automation
  • Advanced scientific and technical analysis (STEM, 3D spatial reasoning)
  • Multimodal RAG on large documents (>200k tokens) and videos

qwen3-vl:30b​

Qwen Team • 30B parameters • Context: 250,000 tokens

State-of-the-art multimodal model (Qwen3-VL) offering exceptional visual understanding and precise temporal reasoning.

Technical specifications:

  • Speed: 43 tokens/second
  • Consumption: 3.10 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ❌ Security

Tags: Vision Agent Large Context Multimodal Video OCR

Use cases:

  • Deep analysis of long videos and intelligent surveillance
  • Extraction of complex structured data (documents, tables, charts)
  • Advanced visual assistants with spatial understanding
  • Multimodal reasoning over sequences of events

qwen3-vl:32b​

Qwen Team • 32B parameters • Context: 250,000 tokens

High-performance variant of Qwen3-VL, optimized for the most demanding vision tasks.

Technical specifications:

  • Speed: 17 tokens/second
  • Consumption: 7.84 kWh/million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ❌ Security

Tags: Vision Agent Large Context Multimodal Video OCR

Use cases:

  • Scientific and technical analysis of high-resolution images
  • Automation of complex visual processes
  • Detailed understanding of dynamic scenes

qwen3:14b​

Qwen Team • 14B parameters • Context: 131,072 tokens

Balanced Qwen3 14B model, delivering strong general performance with good inference speed.

Technical specifications:

  • Speed: 68.2 tokens/second
  • Consumption: 0.90 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Agent Versatile Multilingual

Use cases:

  • High-performance virtual assistants
  • High-quality content generation
  • Classification and extraction tasks

Modelos especializados​

bge-m3:567m​

BAAI • 567M parameters • Context: 8,192 tokens

State-of-the-art multilingual embedding model (BGE-M3), offering exceptional semantic search capabilities across more than 100 languages.

Technical specifications:

  • Speed : 171 tokens/second
  • Energy consumption : 0.36 kWh per million tokens
  • License : MIT
  • Location : FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Embedding Multilingual Efficient

Use cases:

  • Multilingual semantic search
  • Retrieval-Augmented Generation (RAG)
  • Document clustering and classification

deepseek-ocr​

DeepSeek AI • 3B parameters • Context: 8,192 tokens

Specialized OCR model from DeepSeek, designed for high-precision text extraction with formatting preservation.

Technical specifications:

  • Speed: 79 tokens/second
  • Consumption: 1.01 kWh/million tokens
  • License: MIT License
  • Location: FR 🇫🇷

Capabilities: ❌ Tools/Agent • ✅ Vision • ❌ Reasoning • ❌ Security

Tags: Vision OCR Efficient

Use cases:

  • Extraction of structured text (Markdown/LaTeX) from images/PDFs
  • Document digitization with complex tables and formulas

devstral-small-2:24b​

Mistral AI & All Hands AI • 24B parameters • Context: 200,000 tokens

Second iteration of Devstral (Small 2), state-of-the-art agent model for software engineering, deployed on high-performance GPU server.

Technical specifications:

  • Speed: 38 tokens/second
  • Consumption: 3.80 kWh/million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ✅ Security

Tags: Agent Programming Vision Open-Source Large Context FP8 Fast

Use cases:

  • Autonomous coding agents requiring low latency
  • Rapid code refactoring
  • Iterative engineering tasks

devstral:24b​

⚠️ WARNING: This model is deprecated. Recommendation to migrate to devstral-small-2:24b.

Mistral AI & All Hands AI • 24B parameters • Context: 120,000 tokens

Devstral 24b is an agent-based LLM specialized in software engineering, co-developed by Mistral AI and All Hands AI.

Technical specifications:

  • Speed: 44 tokens/second
  • Energy consumption: 3.28 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ✅ Security

Tags: Agent Programming Open-Source Large Context FP8

Use cases:

  • Codebase exploration and modification
  • Autonomous software engineering agents
  • Complex code refactoring and generation

embeddinggemma:300m​

Google • 300M parameters • Context: 2,048 tokens

State-of-the-art embedding model from Google, optimized for its size, ideal for search and semantic retrieval tasks.

Technical specifications:

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Embedding Compact Semantic Efficient Multilingual

Use cases:

  • Information search and retrieval (Retrieval)
  • Document classification and clustering
  • Semantic similarity search
  • Deployment on resource-constrained devices (mobile, laptop)

gemma3:1b​

Google • 1B parameters • Context: 120,000 tokens

Micro-model Gemma 3, ultra-fast and efficient.

Technical specifications:

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Compact Efficient Edge

Use cases:

  • Fast text classification
  • Simple chatbots
  • Rapid prototyping

gemma3:4b​

Google • 4B parámetros • Contexto: 120.000 tokens

Modelo compacto Gemma 3 de 4B, ofreciendo un excelente ratio rendimiento/tamaño.

Especificaciones técnicas:

Capacidades: ❌ Herramientas/Agente • ❌ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Compacto Eficiente Edge

Casos de uso:

  • Asistentes personales en portátiles
  • Resumen de texto
  • Traducción ligera

gpt-oss:20b​

OpenAI • 20B parameters • Context: 120,000 tokens

Open-source language model from OpenAI, optimized for efficiency and deployment on consumer-grade hardware.

Technical specifications:

  • Speed: 9 tokens/second
  • Energy consumption: 14.81 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: MoE Agent Reasoning Open-Source Compact Fast

Use cases:

  • Deployments on resource-constrained devices (edge devices) or low-cost servers.
  • Applications requiring fast inference with strong reasoning capabilities.
  • Agent-based use cases involving function calls, web navigation, and code execution.
  • Fine-tuning for specialized tasks on consumer hardware.

granite-embedding:278m​

IBM • 278 millones de parámetros • Contexto: 8.192 tokens

Modelo de embedding ultra-compacto de IBM Granite, diseñado para una eficiencia máxima.

Especificaciones técnicas:

  • Velocidad: 196,3 tokens por segundo
  • Consumo: 0,31 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ❌ Herramientas/Agente • ❌ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Embedding Compacto Eficiente

Casos de uso:

  • Búsqueda semántica
  • Agrupamiento de documentos

granite4-small-h:32b​

IBM • 32B (9B activos) parámetros • Contexto: 128,000 tokens

Modelo MoE (Mixture-of-Experts) de IBM, diseñado como un "caballo de batalla" para tareas diarias empresariales, con una excelente eficiencia en contextos largos.

Especificaciones técnicas:

  • Velocidad: 49 tokens por segundo
  • Consumo: 2,95 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ❌ Visión • ✅ Razonamiento • ✅ Seguridad

Etiquetas: Agente Razonamiento Seguridad MoE Gran contexto Eficiente

Casos de uso:

  • Agentes conversacionales para soporte al cliente con acceso a bases de conocimientos amplias.
  • Automatización de flujos de trabajo empresariales que requieren el uso de múltiples herramientas.
  • Análisis de documentos largos con un consumo de recursos optimizado.
  • Despliegues en infraestructuras de tamaño medio gracias a su eficiencia.

granite4-tiny-h:7b​

IBM • 7B (1B activos) parámetros • Contexto: 128.000 tokens

Modelo híbrido MoE ultraeficiente de IBM, diseñado para baja latencia, aplicaciones de borde y locales, y como bloque base para flujos de trabajo de agentes.

Especificaciones técnicas:

  • Velocidad: 58 tokens por segundo
  • Consumo: 2,30 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ❌ Visión • ✅ Razonamiento • ✅ Seguridad

Etiquetas: Agente Razonamiento Seguridad MoE Gran contexto Eficiente Rápido Compacto

Casos de uso:

  • Aplicaciones embebidas y de borde que requieren baja latencia.
  • Tareas rápidas dentro de flujos de trabajo de agentes más amplios (por ejemplo: llamadas a funciones).
  • Análisis de documentos en hardware de consumo.
  • Despliegues que requieren una huella de memoria mínima.

medgemma:27b​

Google • 27B parameters • Context: 128,000 tokens

MedGemma is one of Google's most advanced open models for understanding medical text and images, based on Gemma 3.

Technical specifications:

  • Speed: 22 tokens/second
  • Energy consumption: 6.56 kWh per million tokens
  • License: Google Gemma Terms of Use
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ❌ Security

Tags: Medical Vision Specialized Large Context

Use cases:

  • Medical image interpretation (Report generation and VQA)
  • Medical text understanding and clinical reasoning (Decision support)
  • Patient interaction (Interviews and medical triage)
  • Medical record synthesis and literature search

ministral-3:3b​

Mistral AI • 3B parameters • Context: 250,000 tokens

Cutting-edge compact model from Mistral AI, designed for efficiency in local and edge deployments.

Technical specifications:

  • Speed: 50 tokens/second
  • Energy consumption: 1.22 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Compact Efficient Edge

Use cases:

  • Local inference on mobile devices or edge devices
  • Responsive personal assistants
  • Fast routing and classification tasks

ministral-3:8b​

Mistral AI • 8B parameters • Context: 250,000 tokens

Intermediate-sized model from the Ministral family, offering an optimal balance between performance and resource usage.

Technical specifications:

  • Speed: 55 tokens/second
  • Consumption: 2.42 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Efficient Edge Reasoning

Use cases:

  • Advanced local conversational assistants
  • Document analysis and information extraction
  • Tasks requiring a good balance between speed and quality

mistral-small3.2:24b​

Mistral AI • 24B parameters • Context: 128,000 tokens

Minor update to Mistral Small 3.1, improving instruction following, function calling robustness, and reducing repetition errors.

Technical specifications:

  • Speed: 27 tokens/second
  • Energy consumption: 5.35 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ✅ Vision • ❌ Reasoning • ✅ Security

Tags: Vision Agent Security Instruction Following

Use cases:

  • Conversational agents with improved instruction following
  • Robust integration with external tools via function calling
  • Applications requiring high reliability to avoid repetitions
  • Use cases identical to Mistral Small 3.1, with enhanced performance

qwen3-2507-think:4b​

Qwen Team • 4B parameters • Context: 250,000 tokens

Qwen3-4B model optimized for reasoning, with improved performance on logical tasks, mathematics, science, and code, featuring an extended context of 250K tokens.

Technical specifications:

  • Speed: 52 tokens/second
  • Energy consumption: 2.56 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Agent Reasoning Large Context Compact Fast

Use cases:

  • Highly complex reasoning tasks (logic, math, science, code).
  • Conversational agents with very long conversation history (256k tokens).
  • Deep reasoning analysis of very large documents.
  • Integration with external tools via function calling on extremely large contexts.

qwen3-2507:4b​

⚠️ ATENCIÓN: Este modelo está obsoleto. Obsoleto.

Equipo Qwen • 4B parámetros • Contexto: 250.000 tokens

Versión actualizada del modo sin pensamiento de Qwen3-4B, con mejoras significativas en capacidades generales, amplia cobertura de conocimientos y un mejor alineamiento con las preferencias de los usuarios.

Especificaciones técnicas:

  • Velocidad: 30 tokens por segundo
  • Consumo: 4,44 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ❌ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Agente Gran contexto Compacto Rápido Multilingüe

Casos de uso:

  • Tareas generales que requieren un seguimiento preciso de instrucciones y razonamiento lógico.
  • Aplicaciones multilingües con amplia cobertura de conocimientos.
  • Generación de texto de alta calidad para tareas abiertas y subjetivas.
  • Análisis de documentos muy voluminosos gracias al contexto de 256k tokens.

qwen3-embedding:0.6b​

Qwen Team • 0.6B parameters • Context: 32,768 tokens

Ultra-light Qwen3 embedding model, optimized for speed and efficiency on resource-limited infrastructure.

Technical specifications:

  • Speed: N/A
  • Energy consumption: 0.57 kWh per million tokens
  • License: Apache 2.0
  • Localization: FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Embedding Compact Efficient

Use cases:

  • Fast semantic search
  • Real-time text classification

qwen3-embedding:4b​

Qwen Team • 4B parameters • Context: 40,000 tokens

Ultra-performing Qwen3-4B embedding model, offering deep semantic understanding and an extended context window.

Technical specifications:

  • Speed : N/A
  • Energy consumption : 0.57 kWh per million tokens
  • License : Apache 2.0
  • Location : FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Embedding Large Context Efficient

Use cases:

  • Semantic search on long documents
  • RAG with extended context windows
  • High-precision semantic analysis

qwen3-vl:2b​

Qwen Team • 2 millones de parámetros • Contexto: 250.000 tokens

Modelo multimodal ultracompacto Qwen3-VL, que ofrece capacidades avanzadas de visión en dispositivos de borde.

Especificaciones técnicas:

  • Velocidad: 64 tokens por segundo
  • Consumo: 0,95 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ✅ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Visión Compacto Eficiente Multimodal Borde OCR

Casos de uso:

  • Análisis de imágenes en tiempo real en dispositivos móviles
  • OCR y lectura de documentos ligeros
  • Clasificación y ordenación visual rápida

qwen3-vl:4b​

Qwen Team • 4B parámetros • Contexto: 250.000 tokens

Modelo multimodal Qwen3-VL equilibrado, que ofrece un rendimiento sólido en visión con una huella reducida.

Especificaciones técnicas:

  • Velocidad: 57 tokens por segundo
  • Consumo: 2,34 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ✅ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Visión Compacto Multimodal Eficiente Vídeo OCR

Casos de uso:

  • Análisis automatizado de documentos (facturas, formularios)
  • Comprensión de contenido de vídeo
  • Asistentes visuales interactivos

qwen3-vl:8b​

Qwen Team • 8B parámetros • Contexto: 250.000 tokens

Modelo multimodal Qwen3-VL (8B), que ofrece un rendimiento avanzado en visión con una huella razonable.

Especificaciones técnicas:

  • Velocidad: 44 tokens por segundo
  • Consumo: 3,03 kWh por millón de tokens
  • Licencia: Apache 2.0
  • Localización: FR 🇫🇷

Capacidades: ✅ Herramientas/Agente • ✅ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Visión Compacto Multimodal Eficiente Vídeo OCR

Casos de uso:

  • Análisis automatizado de documentos
  • Comprensión de contenido de vídeo
  • Asistentes visuales interactivos

qwen3:0.6b​

Qwen Team • 0.6B parameters • Context: 40,000 tokens

Ultra-light Qwen3 model with 0.6 billion parameters, offering exceptional inference speed for simple and fast tasks.

Technical specifications:

  • Speed: 46 tokens/second
  • Consumption: 1.33 kWh per million tokens
  • License: Apache 2.0
  • Location: FR 🇫🇷

Capabilities: ✅ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Compact Fast Efficient Multilingual

Use cases:

  • Simple text processing tasks
  • Fast classification and sorting
  • Lightweight assistants with low latency

rnj-1:8b​

Essential AI • 8B parameters • Context: 32,000 tokens

8B "Open Weight" model specialized in code, mathematics, and sciences (STEM).

Technical specifications:

  • Speed: 31 tokens/second
  • Consumption: 1.97 kWh/million tokens
  • License: Open Weights
  • Location: FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ✅ Reasoning • ❌ Security

Tags: Code Maths STEM Reasoning Efficient

Use cases:

  • Advanced programming assistant and code generation
  • Solving complex mathematical problems
  • Scientific and technical tasks (STEM)

translategemma:12b​

Google • 12B parameters • Context: 128,000 tokens

State-of-the-art open translation model based on Gemma 3, covering 55 languages.

Technical specifications:

  • Speed : 30 tokens/second
  • Consumption : 4.44 kWh/million tokens
  • License : Gemma Terms of Use
  • Location : FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Translation Multilingual Specialized

Use cases:

  • Translation of long documents
  • Inter-language communication
  • Content localization

translategemma:27b​

Google • 27B parámetros • Contexto: 120.000 tokens

Modelo de traducción de alto rendimiento basado en Gemma 3 27B.

Especificaciones técnicas:

  • Velocidad: 44 tokens por segundo
  • Consumo: 6,35 kWh por millón de tokens
  • Licencia: Términos de uso de Gemma
  • Localización: FR 🇫🇷

Capacidades: ❌ Herramientas/Agente • ❌ Visión • ❌ Razonamiento • ❌ Seguridad

Etiquetas: Traducción Multilingüe Especializado Alto rendimiento

Casos de uso:

  • Traducción de alta precisión
  • Traducción de documentos técnicos
  • Matrices literarias y culturales

translategemma:4b​

Google • 4B parameters • Context: 128,000 tokens

Compact version of the TranslateGemma translation model, optimized for speed.

Technical specifications:

  • Speed : 38 tokens/second
  • Consumption : 1.27 kWh per million tokens
  • License : Gemma Terms of Use
  • Location : FR 🇫🇷

Capabilities: ❌ Tools/Agent • ❌ Vision • ❌ Reasoning • ❌ Security

Tags: Translation Multilingual Specialized Efficient

Use cases:

  • Fast text translation
  • Translation on devices with limited resources
  • Real-time localization

Cas de uso recomendados​

Multilingual dialogue​

Chatbots and assistants capable of communicating in multiple languages with automatic language detection, context preservation throughout the conversation, and understanding of linguistic nuances.

Recommended models:

  • Llama 3.3
  • Mistral Small 3.2
  • Qwen 3
  • Openai OSS
  • Granite 4

Análisis de documentos largos​

Procesamiento de documentos extensos (>100 páginas) con mantenimiento del contexto a lo largo de todo el texto, extracción de información clave, generación de resúmenes relevantes y respuesta a preguntas específicas sobre el contenido.

Modelos recomendados:

  • Gemma 3
  • Qwen next
  • Qwen 3
  • Granite 4

Programación y desarrollo​

Generación y optimización de código en múltiples lenguajes, depuración, refactorización, desarrollo de funcionalidades completas, comprensión de implementaciones algorítmicas complejas y creación de pruebas unitarias

Modelos recomendados:

  • DeepCoder
  • Qwen3 coder
  • Granite 4
  • Devstral

Visual analysis​

Direct processing of images and visual documents without prior OCR preprocessing, interpretation of technical diagrams, charts, tables, drawings, and photos, with generation of detailed textual explanations of the visual content.

Recommended models:

  • deepseek-OCR
  • Mistral Small 3.2
  • Gemma 3
  • Qwen 3 VL

Seguridad y cumplimiento​

Aplicaciones que requieren capacidades específicas en materia de seguridad; filtrado de contenido sensible, trazabilidad de razonamientos, verificación del RGPD/HDS, minimización de riesgos, análisis de vulnerabilidades y cumplimiento de regulaciones sectoriales.

Modelos recomendados:

  • Granite Guardian
  • Granite 4
  • Devstral
  • Mistral Small 3.2
  • Magistral small

Lightweight and Embedded Deployments​

Applications requiring minimal resource footprint, deployment on devices with limited capacity, real-time inference on standard CPUs, and integration into embedded systems or IoT devices

Recommended models:

  • Gemma 3n
  • Granite 4 tiny
  • Qwen 3 VL (2B)