Modèles d’IA gratuits sur OpenRouter

Classement des modèles mis à jour à partir de données d’utilisation réelles.

Les modèles sont classés selon le nombre total de jetons de prompt et de réponse traités via l’API OpenRouter au cours des 7 derniers jours. Affichage des 16 meilleurs modèles. Les classements mesurent l’utilisation sur OpenRouter et reflètent l’adoption, et non la qualité des modèles ou leurs performances aux tests de référence.

Les modèles d’IA gratuits les plus utilisés sont classés ci-dessous selon leur utilisation réelle sur OpenRouter au cours de la semaine passée, et tous ont un tarif de 0 $ pour les prompts. Les meilleurs modèles actuels sont Nemotron 3 Ultra (free), Nemotron 3.5 Super VL (free) et ERNIE-5.5-Flash-Preview (free). Les modèles gratuits vous permettent de créer des prototypes, d’apprendre et de mener des projets personnels avec un budget quasi nul, et OpenRouter continue d’accroître la capacité gratuite en intégrant des fournisseurs et en prenant directement en charge les coûts. Si vous préférez ne pas choisir, openrouter/free achemine chaque requête vers un modèle gratuit adapté.

Meilleurs modèles gratuits sur OpenRouter

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it supports text input and output with a context window of up to 1M tokens. It is suited for long-running agentic workflows, including agent orchestration, coding agents, deep research, and complex enterprise tasks.

It is particularly strong at multi-step reasoning and planning, with high-throughput inference designed for high-volume agent pipelines. It is part of the NVIDIA Nemotron family of open models for agentic AI.

par nvidia1M de contexte0 $US/M entrée 0 $US/M sortie

NVIDIA Nemotron 3.5 Super VL is an open multimodal mixture-of-experts model from NVIDIA, with 12B active parameters out of 120B total. It takes text, images, documents, GUI screenshots, and video as input and returns text, and is built for long-running agents that combine visual understanding with reasoning and tool use.

It uses a hybrid Mamba-Transformer architecture with multi-token prediction and supports a 1M-token context. Typical uses include document intelligence, web research, coding validation, customer service, and video understanding. Reasoning is on by default and can be turned off or given a token budget.

par nvidia1M de contexte0 $US/M entrée 0 $US/M sortie

ERNIE 5.5 Flash Preview is the latest foundation model in the ERNIE family, with comprehensive improvements in coding, agentic capabilities, knowledge work, and deep research. It has 280 billion total parameters, activates 14 billion parameters per token, and supports a context window of 1 million tokens. Its new architecture combines strong model capabilities with efficient inference, delivering generation speeds of more than 200 tokens per second.

par baidu1,02M de contexte0 $US/M entrée 0 $US/M sortie

NVIDIA Nemotron 3.5 Lightning is an open mixture-of-experts model from NVIDIA, with 3B active parameters out of 30B total. It is suited for high-throughput agentic workloads and specialized tasks that benefit from domain-specific customization.

par nvidia1M de contexte0 $US/M entrée 0 $US/M sortie

Inkling 2 Small is an open-weight multimodal mixture-of-experts reasoning model from Thinking Machines Lab, with 17B active parameters out of 594B total. It is the first model in the second Inkling generation. It was trained with reinforcement learning on long-horizon coding tasks across many agent harnesses, and is suited for agentic coding, autonomous research, agentic knowledge work and retrieval-augmented generation, with English and general multilingual support. It accepts text and image input (no audio), supports tool calling and variable reasoning effort, and has a 1M-token context window.

par thinkingmachines1,05M de contexte0 $US/M entrée 0 $US/M sortie

Dots3-Note Preview is an open-weight mixture-of-experts model from Dots Studio, with 16B active parameters out of 280B total. It is the lightest model in the Dots 3 family and is suited for reasoning, coding, multimodal understanding, long-context processing, and multi-step agent workflows.

par dots-studio512K de contexte0 $US/M entrée 0 $US/M sortie

Particle-1 is a token-efficient agentic coding model from Particle AI, post-trained on DeepSeek V4.1 Flash. It is suited for software engineering work such as terminal workflows, codebase changes, and data engineering tasks, and is trained to complete them with substantially fewer reasoning and tool-call tokens while keeping accuracy close to its base model.

par particle-ai1,05M de contexte0 $US/M entrée 0 $US/M sortie

North Mini Code is Cohere's first agentic coding model and the debut of its North family. A sparse mixture-of-experts model with 30B total parameters and 3B active, it is optimized for code generation, agentic software engineering, and terminal tasks, and is trained to generalize across agent harnesses such as OpenCode and SWE-Agent.

It offers a 256K-token context window with up to 64K tokens of output, supports interleaved reasoning and tool use via JSON schema, and is released open-weight under the Apache 2.0 license. Its small active-parameter footprint enables low-latency inference, including on local hardware.

par cohere256K de contexte0 $US/M entrée 0 $US/M sortie

Flint is an open-weight mixture-of-experts model from Recursive, with about 2.5B active parameters out of 40B total, built by Recursive's autoresearch system Eureka Machine for coding agents. It is suited for investigating codebases, fixing bugs, implementing features, and writing tests.

par recursive262K de contexte0 $US/M entrée 0 $US/M sortie

Ling 3.2 Flash is a hybrid reasoning model from inclusionAI that accepts text and image input, with a 1M-token context window and up to 200 images per request. Thinking can be switched off per request.

par inclusionai1,05M de contexte0 $US/M entrée 0 $US/M sortie

Decider V1.3 is a decision model from Perplexity, succeeding Decider V1.2. Instead of generating text, it reads content passed as state (text, JSON, or images) and returns typed, probabilistic answers to one or more named questions in a single request: the probability of yes for a yes/no question (noul), a probability for every option plus the most likely one (choice, up to 255 options), or a probability for every level of an ordered rubric plus the expected score (score).

It is built for classification, routing, moderation, and rubric grading where application code thresholds the returned numbers rather than parsing a chat reply.

V1.3 also adds locate, served on the Extractions route: it pulls text and list fields out of text state and returns each value with its source span and a confidence. For a smaller, lower-cost option with the same contract, see Decider Flash V1.3.

par perplexity0 $US/M entrée 0 $US/M sortie

NVIDIA Nemotron™ 3 Nano Omni is a 30B-A3B open multimodal model designed to function as a perception and context sub-agent in enterprise agent systems. It accepts text, image, video, and audio inputs and produces text output, enabling agents to perceive and reason across modalities in a single inference loop.

Built on a hybrid MoE Transformer-Mamba architecture with Conv3D video layers and Efficient Video Sampling (EVS), it delivers approximately 2× higher throughput and 2.5× lower compute for video reasoning versus separate vision + speech pipelines. It supports up to 300K context length and a 16,384 reasoning budget, with extended thinking enabled via reasoning.enabled on OpenRouter.

par nvidia256K de contexte0 $US/M entrée 0 $US/M sortie

A compact 1.7B-parameter text model with a 256K-token context, post-trained for agents, tool calling, and structured outputs. It reasons before it answers. Prompts and outputs may be retained and used to train Liquid models.

par liquid262K de contexte0 $US/M entrée 0 $US/M sortie

Decider Flash V1.3 is a decision model from Perplexity, a smaller, lower-cost sibling of Decider V1.3 with the same API contract. Instead of generating text, it reads content passed as state (text, JSON, or images) and returns typed, probabilistic answers to one or more named questions in a single request: the probability of yes for a yes/no question (noul), a probability for every option plus the most likely one (choice, up to 255 options), or a probability for every level of an ordered rubric plus the expected score (score).

It suits high-volume classification, routing, and moderation where application code thresholds the returned numbers rather than parsing a chat reply.

It also adds locate, served on the Extractions route, pulling text and list fields out of text state with source spans and confidences.

par perplexity0 $US/M entrée 0 $US/M sortie

Span-01 Lite is the free, lighter tier of Span-01, a behavior scoring model from Respan. It returns, for each plain-language behavior you define, the probability that the behavior is present in a conversation span, and is suited for high-volume evaluation and monitoring where cost matters more than peak accuracy.

par respan0 $US/M entrée 0 $US/M sortie

Llama Nemotron Rerank VL 1B V2 is a 1.7B multimodal reranking model from NVIDIA. It evaluates the relevance of document images and text against user queries, designed for vision RAG pipelines handling charts, tables, infographics, and mixed-media documents. Functions as a cross-encoder that accepts text queries paired with image, text, or combined document inputs, delivering approximately 6-7% recall improvements over embedding-only baselines on visual document retrieval benchmarks.

par nvidia0 $US/M entrée 0 $US/M sortie

Questions fréquentes

Explorer d’autres collections