Repli de fournisseur
OpenRouter peut automatiquement rerouteur une requête échouée vers un fournisseur alternatif pour le même modèle, réduisant les défaillances graves pour les appels LLM.
OpenRouter lets you swap the model behind Claude Code and other tools by changing ANTHROPIC_BASE_URL — 300+ language models, some with free tiers. RunAPI does the same base URL swap, adds image, video, music, and audio (210+ models total), and charges 50% off official rates with no deposit required to avoid rate limits. Both expose an OpenAI-compatible API.
OpenRouter is a well-known LLM gateway with 49,500 monthly searches and domain rating 85. It routes 300+ language models, including free tiers on select community models, and includes provider fallback so a failed call can retry against another provider. It does not host models itself — it reroutes requests. RunAPI covers the same LLM families (Claude, GPT, Gemini) plus image, video, music, and audio under one key, charges 50% off official rates with no deposit or request cap, and ships an MCP server, CLI, and typed SDKs.
Both expose an OpenAI-compatible API. Changing the base URL and swapping the key is usually the only change required.
OpenRouter est uniquement LLM. RunAPI ajoute la génération d'images, de vidéos, de musique et d'audio sous la même clé.
OpenRouter’s free tier caps at 50 requests per day without a $10 deposit. RunAPI is pay-as-you-go with no free tier and no request cap.
RunAPI propose un serveur MCP, une CLI, des SDK pour Python, Node.js, PHP, Java, Go et Ruby, et une bibliothèque de prompts consultable.
The table compares the two platforms on the dimensions developers weigh when choosing an API gateway: modality, model count, tooling, data handling, and pricing.
| Fonctionnalité | OpenRouter | RunAPI |
|---|---|---|
| Routage LLM | Oui — 300+ modèles | Oui |
| Génération d'images | Non | Oui |
| Génération de vidéos | Non | Oui |
| Musique et audio | Non | Oui |
| Niveau de modèles gratuits | Oui — modèles sélectionnés | Non |
| Serveur MCP | Non | Oui |
| SDK officiels | Compatible OpenAI uniquement | Python, Node.js, PHP, Java, Go, Ruby + compatible OpenAI |
| Outil CLI | Non | Oui |
| Compétences d'agent | Non | Oui |
| Bibliothèque de prompts | Non | Oui |
| API compatible OpenAI | Oui | Oui |
| Tarification LLM | Tarifs fournisseur + majoration | 50 % de réduction sur les tarifs officiels |
OpenRouter has been running since 2023 and has a large developer community. It is a rerouter — it does not host models itself — so uptime depends on both OpenRouter’s own infrastructure and the providers it routes to. Free-tier models are rate-limited and queue behind paid traffic, which is the single biggest complaint from Claude Code users who hit the 50 req/day cap on a $0-balance account. Reliability considerations worth knowing before depending on OpenRouter as your production LLM gateway are below.
OpenRouter peut automatiquement rerouteur une requête échouée vers un fournisseur alternatif pour le même modèle, réduisant les défaillances graves pour les appels LLM.
Free-tier models are rate-limited to 50 requests per day on a $0 balance, or 1,000 requests per day after you deposit $10. Production workloads that burn through those caps will hit queuing delays behind paid traffic.
OpenRouter route uniquement les modèles de langage. Les équipes ayant besoin de génération d'images, de vidéos ou de musique doivent ajouter un fournisseur séparé, ce qui signifie gérer deux passerelles.
OpenRouter publie une page de statut. Comme tout proxy tiers, sa disponibilité dépend de sa propre infrastructure et des fournisseurs en amont qu'il route. Consultez la page de statut pour l'historique des incidents actuels.
OpenRouter focuses on language models at volume, including community and open-source models — it is a rerouter, not a model host, so uptime depends on the underlying providers it routes to. RunAPI routes commercial production models and adds generative media categories that OpenRouter does not cover. Your requests go to the same frontier providers either way; the difference is what categories and tooling sit on top.
Claude, GPT, and Gemini families for chat, reasoning, coding, and tool use — the same LLM models OpenRouter users expect, at 50% off official rates.
Modèles texte-vers-image et édition d'image incluant des variantes Flux et d'autres points de terminaison de génération, appelables avec la même clé API.
Génération texte-vers-vidéo et image-vers-vidéo, facturée à l'utilisation par tâche.
Génération de musique et synthèse audio pour les bandes sonores, les voix off et les effets — des modalités qu'OpenRouter ne prend pas en charge.
OpenRouter passes through provider pricing and adds a markup on paid models. Free-tier models are rate-limited: accounts with no balance get 50 requests per day; depositing $10 raises that cap to 1,000. RunAPI has no free inference tier and no deposit requirement — you pay 50% of each provider’s official published rate from the first token. The table shows representative flagship LLM rates per million tokens.
| Modèle | Entrée officielle /M | Sortie officielle /M | Entrée RunAPI /M | Sortie RunAPI /M |
|---|---|---|---|---|
| Claude Sonnet 4.6 | $6.00 | $30.00 | $3.00 | $15.00 |
| Claude Opus 4.7 | $10.00 | $50.00 | $5.00 | $25.00 |
| GPT-5.4 | $2.50 | $15.00 | $1.25 | $7.50 |
| Gemini 2.5 Pro | $1.25 | $10.00 | $0.63 | $5.00 |
RunAPI applique une réduction de 50 % sur tous les fournisseurs. Les modèles médias sont facturés par tâche. Tarifs vérifiés en juin 2026.
Inscrivez-vous sur runapi.ai. Aucune carte de crédit requise pour commencer.
Allez dans Dashboard → API Keys, créez une clé et enregistrez-la dans votre environnement.
Change your client’s base URL to https://runapi.ai/v1 and use your RunAPI key. If you’re using Claude Code, set ANTHROPIC_BASE_URL to https://runapi.ai/v1 and ANTHROPIC_API_KEY to your RunAPI key in settings.json. OpenAI-compatible code requires no other changes.
Accédez aux modèles d'image, de vidéo ou d'audio via la même clé, ou installez le serveur MCP et la CLI pour les workflows non-code.
RunAPI does not train on your requests and does not log prompt content for model improvement. Free models on OpenRouter are hosted by third-party providers under OpenRouter’s terms — those providers’ data policies apply. If you are routing client data or sensitive code, check the data policy of every provider in the chain, not just the gateway.
OpenRouter throttles accounts with $0 balance to 50 requests per day on free models. Depositing $10 raises that to 1,000 requests per day. RunAPI has no free inference tier, so there is no deposit requirement or request-cap — you pay per token at 50% of official rates from the first call.
Free models on OpenRouter are rate-limited and queue behind paid traffic. When you pin to a single free model you can hit the cap fast; the openrouter/auto endpoint spreads load across available free models to reduce that. RunAPI has no free tier, so all paid requests go through without queuing behind free traffic.
Yes. Set ANTHROPIC_BASE_URL to https://runapi.ai/v1 and use your RunAPI key as ANTHROPIC_API_KEY. The Claude Code shell — its agentic framework, slash commands, and tool calls — stays the same. You are changing the model that processes the prompts, not the client itself.
No. Both platforms expose an OpenAI-compatible API, so switching is a base URL and key change in your existing client. Point your client at https://runapi.ai/v1, swap in your RunAPI key, and keep the rest of your code as-is. You can run both in parallel during testing.
RunAPI charges 50% of each provider's official published rate. Claude Sonnet 4.6 runs $3 input and $15 output per million tokens versus $6 and $30 official; GPT-5.4 runs $1.25 input and $7.50 output versus $2.50 and $15. Image, video, and audio are billed per task at the same half-off positioning.
RunAPI ships an MCP server for MCP-aware tools like Claude Code and Cursor, a command-line tool, typed SDKs for Python, Node.js, PHP, Java, Go, and Ruby, installable agent skills, and a searchable prompt library. OpenRouter provides an OpenAI-compatible API and a web UI but does not include these additional tooling layers.
RunAPI covers LLMs from Claude, GPT, and Gemini families plus image, video, music, and audio generation — 210+ models in total. OpenRouter routes 300+ language models including many open-source and community models, but covers text only. If you need free or open-source LLM access at volume, OpenRouter has more options there.
RunAPI vs Requesty : tarifs, fonctionnalités et selection de modèles.
API hébergée vs proxy auto-hébergé — quand choisir RunAPI plutôt que LiteLLM.
Tarifs GPT-5.4, GPT-5.4-mini et o3 : RunAPI vs tarifs officiels.
Parcourez 210+ modèles avec tarifs, paramètres et exemples de code.
Créez un compte RunAPI gratuit pour 210+ modèles en texte, image, vidéo et audio — compatible OpenAI et 50 % de réduction sur les tarifs officiels.