Aller au contenu
Alpha
  • Modèles
  • Classements
  • Ori
Contacter l’équipe commerciale

Pied de page

OpenRouterOpenRouter
© 2026 OpenRouter, Inc

Produit

  • Chat
  • Classements
  • Évaluations comparatives
  • Applications
  • Découvrir
  • Modèles
  • Ori
  • Collections
  • Fournisseurs
  • Outils
  • Tarifs
  • Business
  • Enterprise
  • Laboratoire

Entreprise

  • À propos
  • Blog
  • Clients
  • Carrières
    Recrutement
  • Confidentialité
  • Conditions d’utilisation
  • Centre de confiance
  • Assistance
  • Compatible avec OR
  • Données
  • Marque

Développeur

  • Documentation
  • Référence de l’API
  • Plateforme pour développeurs
  • Statut
  • Plan du site pour l’IA

Connecter

  • Discord
  • GitHub
  • LinkedIn
  • X
  • YouTube
Favicon de anthropic

Anthropic: Claude Haiku 4.5

anthropic/claude-haiku-4.5

Comparer

Claude Haiku 4.5 is Anthropic’s fastest and most efficient model, delivering near-frontier intelligence at a fraction of the cost and latency of larger Claude models. Matching Claude Sonnet 4’s performance across reasoning, coding, and computer-use tasks, Haiku 4.5 brings frontier-level capability to real-time and high-volume applications.

It introduces extended thinking to the Haiku line; enabling controllable reasoning depth, summarized or interleaved thought output, and tool-assisted workflows with full support for coding, bash, web search, and computer-use tools. Scoring >73% on SWE-bench Verified, Haiku 4.5 ranks among the world’s best coding models while maintaining exceptional responsiveness for sub-agents, parallelized execution, and scaled deployment.

Modalités
Prix en entrée / sortie
1 $US / 5 $USpar 1M
Contexte
200K
Publié
15 oct. 2025
Comparer
FournisseursTarificationPerformancesDisponibilitéÉvaluations comparativesApplicationsActivitéFAQExplorer

Fournisseurs

Différentes entreprises hébergent le même modèle. OpenRouter achemine votre requête vers l’une d’elles selon le mode de routage choisi — Balanced (prix + vitesse), Nitro (le plus rapide) ou Floor (le moins cher).

Tarification

Le prix moyen que les clients paient réellement pour ce modèle, à côté des prix affichés par les fournisseurs. Grâce à la mise en cache et aux remises, le prix réellement payé est souvent bien inférieur au prix affiché.

Performances

Le débit est la vitesse à laquelle le modèle écrit (jetons par seconde — plus il est élevé, mieux c’est). La latence est le temps total d’aller-retour (plus il est faible, mieux c’est). Le TTFT est le délai avant le premier jeton — le temps avant de voir quelque chose apparaître (plus il est faible, mieux c’est).

Disponibilité

La disponibilité correspond au pourcentage des 3 derniers jours pendant lequel au moins un fournisseur répondait aux requêtes. La disponibilité du service correspond au pourcentage de temps pendant lequel l’inférence a été assurée avec succès. OpenRouter surveille en continu les fournisseurs et utilise le meilleur fournisseur suivant lorsque l’un d’eux renvoie une erreur.

Évaluations comparatives

Scores aux évaluations standardisées et réponses de ce modèle aux prompts communs des tests de référence. Des pourcentages plus élevés sont meilleurs — et chaque rang indique la position de ce modèle parmi les modèles évalués sur cette source.

Résumé des scores aux tests de performance pour Anthropic: Claude Haiku 4.5 (Artificial Analysis, Design Arena et Vals AI)
SourceTest de performanceScore
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) GPQA Diamond64.6%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) HLE4.2%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) IFBench42.0%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) τ²-Bench Telecom32.5%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) AA-LCR49.7%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) CritPt0.0%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) LiveCodeBench51.1%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) Terminal-Bench Hard27.3%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) MMLU-Pro80.0%
Artificial AnalysisClaude 4.5 Haiku (Non-reasoning) AIME 202539.0%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) Coding Index43,9
Artificial AnalysisClaude 4.5 Haiku (Reasoning) Agentic Index8,0
Artificial AnalysisClaude 4.5 Haiku (Reasoning) GPQA Diamond67.2%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) HLE10.4%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) IFBench54.3%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) τ²-Bench Telecom54.7%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) AA-LCR74.3%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) τ-Bench Banking9.3%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) GDPval-AA12.0%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) CritPt0.0%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) LiveCodeBench61.5%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) SciCode42.2%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) Terminal-Bench Hard27.3%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) Terminal-Bench 4.00.0%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) Terminal-Bench 2.144.2%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) MMLU-Pro76.0%
Artificial AnalysisClaude 4.5 Haiku (Reasoning) AIME 202583.7%
Design ArenaClaude Haiku 4.5 Models Arena 3D Elo1069
Design ArenaClaude Haiku 4.5 Models Arena Asciiart Elo1137
Design ArenaClaude Haiku 4.5 Models Arena Code Categories Elo1113
Design ArenaClaude Haiku 4.5 Models Arena Data Visualization Elo1125
Design ArenaClaude Haiku 4.5 Models Arena Game Development Elo1097
Design ArenaClaude Haiku 4.5 Models Arena SVG Elo1022
Design ArenaClaude Haiku 4.5 Models Arena UI Component Elo1095
Design ArenaClaude Haiku 4.5 Models Arena Website Elo1122
Vals AIClaude Haiku 4.5 (Nonthinking) CorpFin v260.3%
Vals AIClaude Haiku 4.5 (Nonthinking) Public Benefits Bench v1.154.3%

Applications

Applications publiques qui envoient le plus de trafic à ce modèle. Un bon indicateur des charges de travail réelles en production — et un aperçu des cas d’usage auxquels ce modèle convient le mieux.

Activité

Volume de jetons et trafic des requêtes vers ce modèle au fil du temps.

API

Code prêt à l’emploi pour appeler ce modèle. L’API d’OpenRouter est compatible avec OpenAI — la plupart des SDK fonctionnent en remplaçant simplement l’URL de base. Seul le slug du modèle ci-dessous change d’un modèle à l’autre.

Versions de Anthropic: Claude Haiku

Disponible via le routage au sein de la région UE

Anthropic: Claude Haiku 4.5 est disponible via le routage intrarégional dans l’UE. Les requêtes envoyées à eu.openrouter.ai sont déchiffrées dans l’UE et acheminées uniquement vers des points de terminaison partagés de fournisseurs en Europe (UE/EEE, Royaume-Uni et Suisse).

Fournisseurs en Europe
Amazon Bedrock, Google Vertex
Tarifs en Europe
  • Amazon Bedrock : 1,10 $US/M tokens en entrée, 5,50 $US/M tokens en sortie
  • Google Vertex : 1,10 $US/M tokens en entrée, 5,50 $US/M tokens en sortie
URL de base
https://eu.openrouter.ai/api/v1
Offre
Nécessite Business ou Enterprise. En libre-service avec Business : sans contrat ni minimum mensuel. OpenRouter Business pour le routage au sein de l’UE

Blocage en cas d’échec : si aucun point de terminaison de fournisseur partagé en Europe n’est disponible, la requête renvoie une erreur au lieu de quitter la région. Les requêtes traitées avec votre propre clé BYOK sont acheminées vers la région de votre déploiement, que vous contrôlez. Documentation sur le routage intrarégional

Explorer d’autres modèles

Modèles d’IA avec vision : LLM multimodaux pour la compréhension d’imagesCollectionClassements des modèles d’IAClassement
Latence / débit
1,00 $US5,00 $US0,10 $US0,39s70 tps
100,00 %
1,00 $US5,00 $US0,10 $US0,70s103 tps
100,00 %
1,00 $US5,00 $US0,10 $US0,60s85 tps
99,77 %
1,00 $US5,00 $US0,10 $US0,45s65 tps
100,00 %
Non utilisé pour le routage Standard :Pourquoi ces points de terminaison ne sont pas utilisés
1,10 $US5,50 $US0,11 $US0,47s125 tps
100,00 %
1,10 $US5,50 $US0,11 $US2,56s187 tps
100,00 %
1,10 $US5,50 $US0,11 $US0,47s171 tps
100,00 %
1,10 $US5,50 $US0,11 $US0,29s93 tps
100,00 %
Débit
187jetons/s

P50, meilleur résultat tous fournisseurs confondus

Latence
0,29s

P50, meilleur fournisseur

Évaluations comparatives AutoExacto
GPQA DiamondTAU-BenchGoogle Vertex72,9 %67,1 %Amazon Bedrock (EU)72,7 %67,2 %Amazon Bedrock (US)73,2 %66,4 %auto-routing73,0 %65,9 %Anthropic72,5 %65,8 %Azure72,9 %65,1 %Google Vertex (EU)71,3 %66,7 %Amazon Bedrock72,2 %65,5 %Google Vertex (US)72,3 %64,5 %
Disponibilité (3 j)Le modèle était joignable. Requête acheminée vers un fournisseur.
100,00 %
Disponibilité (3 j)Le modèle a renvoyé un résultat d’inférence, quel que soit le fournisseur. Les erreurs et les réponses vides sont comptabilisées à son détriment.
99,87 %

Disponibilité au cours des 3 derniers jours

72 dernières heures
Disponibilité 99,87 %
Il y a 3 joursIl y a 2 joursHierMaintenant

Disponibilité au cours des dernières 24 heures

OpenRouter Availability
99,88 %
Without Routing
98,69 %

Lorsqu’une erreur survient chez un fournisseur en amont, nous pouvons rétablir le service en routant la requête vers un autre fournisseur opérationnel, si les filtres de votre requête le permettent. Vous pouvez accéder par programmation aux données de disponibilité de chaque fournisseur via l’ API des points de terminaison. Découvrez nos options d’ équilibrage de charge et de routage des fournisseurs.

1.
Favicon de https://claude.ai/apple-touch-icon.png
Claude Code
Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.
48,3Bjetons
2.
Favicon de https://nousresearch.com
Hermes Agent
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
25,1Bjetons
3.
Favicon de https://descript.com/
Descript
AI Video & Podcast Editor
15,1Bjetons
4.
Favicon de https://openclaw.ai/
OpenClaw
OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from running commands and browsing the web to managing files and sending emails.
14,2Bjetons
5.
Favicon de https://revring.ai/
RevRing Max
nouveau
13,9Bjetons

Questions fréquentes

Claude Haiku 4.5 is Anthropic’s fastest and most efficient model, delivering near-frontier intelligence at a fraction of the cost and latency of larger Claude models. Matching Claude Sonnet 4’s performance across reasoning, coding, and computer-use tasks, Haiku 4.5 brings frontier-level capability to real-time and high-volume applications.

Claude Haiku 4.5 coûte 1,00 $US/M de jetons d’entrée et 5,00 $US/M de jetons de sortie, avec des tarifs distincts pour Cache Read à 0,10 $US/M tokens, Cache Write à 1,25 $US/M tokens et Cache Write (1h) à 2,00 $US/M tokens.

Claude Haiku 4.5 dispose d’une fenêtre de contexte de 200 000 jetons. Il prend en charge jusqu’à 64 000 jetons de complétion.

Oui. Claude Haiku 4.5 accepte tools et tool_choice pour l’appel de fonctions. Il prend également en charge les sorties structurées via un schéma JSON dans response_format.

Claude Haiku 4.5 accepte texte, images et fichiers tels que des PDF en entrée et renvoie texte.

Claude Haiku 4.5 est proposé par 4 fournisseurs sur OpenRouter : Azure, Amazon Bedrock, Google Vertex et Anthropic. Les requêtes sont acheminées vers le meilleur fournisseur disponible, avec un basculement automatique vers les autres, et vous pouvez fixer ou exclure des fournisseurs avec le routage des fournisseurs.

Claude Haiku 4.5 est sorti le 15 octobre 2025.

Autres modèles de Anthropic

Claude Haiku 5.5

Claude Haiku 5.5 is Anthropic's small, fast model for high-volume, cost-sensitive work such as summarization, subagents, and browser use. It succeeds Claude Haiku 4.5 with stronger coding, computer use, and knowledge work, and accepts text and image input with a 1M-token context window.

It is the first Haiku model with adjustable effort. Thinking is adaptive and on by default, with effort as the main lever for trading off depth, latency, and cost; it can be turned off at low, medium, and high effort.

TexteContexte de 1M0,10 $US / 0,50 $US
Claude Haiku 5.5 (batch)

Claude Haiku 5.5 is Anthropic's small, fast model for high-volume, cost-sensitive work such as summarization, subagents, and browser use. It succeeds Claude Haiku 4.5 with stronger coding, computer use, and knowledge work, and accepts text and image input with a 1M-token context window.

It is the first Haiku model with adjustable effort. Thinking is adaptive and on by default, with effort as the main lever for trading off depth, latency, and cost; it can be turned off at low, medium, and high effort.

TexteContexte de 1M0,05 $US / 0,25 $US
Claude Sonnet 5.5

Claude Sonnet 5.5 is Anthropic's Sonnet-class model for well-scoped everyday work, succeeding Claude Sonnet 5 as a direct upgrade. It is especially strong at building features, fixing bugs, and producing polished documents, slides, and spreadsheets, and it writes and communicates more clearly than its predecessor.

Thinking is always on, so effort is the main lever for trading off depth, latency, and cost, and lower effort settings keep it responsive for everyday agentic loops.

TexteContexte de 1M2 $US / 10 $US
Claude Sonnet 5.5 (batch)

Claude Sonnet 5.5 is Anthropic's Sonnet-class model for well-scoped everyday work, succeeding Claude Sonnet 5 as a direct upgrade. It is especially strong at building features, fixing bugs, and producing polished documents, slides, and spreadsheets, and it writes and communicates more clearly than its predecessor.

Thinking is always on, so effort is the main lever for trading off depth, latency, and cost, and lower effort settings keep it responsive for everyday agentic loops.

TexteContexte de 1M1 $US / 5 $US
Claude Opus 5.5

Claude Opus 5.5 is Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work, succeeding Claude Opus 5. It is particularly strong at multi-step changes in large codebases, code review and bug finding, financial and scientific analysis, and reading dense charts, diagrams, and screenshots, and it is more careful than its predecessor about only stating figures and citing sources it can back up.

The model completes comparable tasks in fewer steps and with fewer tokens than Opus 5, and reports on its work in plainer language, with clear updates on what it did, what it found, and what it needs from the user. Thinking is always adaptive, so effort is the main lever for trading off depth, latency, and cost, and lower effort settings remain effective for latency-sensitive workloads.

TexteContexte de 1M4 $US / 20 $US
Claude Opus 5.5 (batch)

Claude Opus 5.5 is Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work, succeeding Claude Opus 5. It is particularly strong at multi-step changes in large codebases, code review and bug finding, financial and scientific analysis, and reading dense charts, diagrams, and screenshots, and it is more careful than its predecessor about only stating figures and citing sources it can back up.

The model completes comparable tasks in fewer steps and with fewer tokens than Opus 5, and reports on its work in plainer language, with clear updates on what it did, what it found, and what it needs from the user. Thinking is always adaptive, so effort is the main lever for trading off depth, latency, and cost, and lower effort settings remain effective for latency-sensitive workloads.

TexteContexte de 1M2 $US / 10 $US
Claude Fable 5.1

Claude Fable 5.1 improves on Claude Fable 5 across the board, with the biggest gains in agentic coding, long-running agentic workflows, and knowledge work: long code refactors, front-end and visual code generation, and finance and analysis tasks in particular. It also tends to be more concise than Fable 5 in its plans and summaries. We recommend testing it as a direct upgrade wherever you use Fable 5 today, and alongside Opus 5 on reasoning-heavy tasks.

TexteContexte de 1M10 $US / 50 $US
Claude Fable 5.1 (batch)

Claude Fable 5.1 improves on Claude Fable 5 across the board, with the biggest gains in agentic coding, long-running agentic workflows, and knowledge work: long code refactors, front-end and visual code generation, and finance and analysis tasks in particular. It also tends to be more concise than Fable 5 in its plans and summaries. We recommend testing it as a direct upgrade wherever you use Fable 5 today, and alongside Opus 5 on reasoning-heavy tasks.

TexteContexte de 1M5 $US / 25 $US
Claude Opus 5

Fast-mode variant of Opus 5 - identical capabilities with higher output speed at 2x pricing relative to regular Opus 5.

Learn more in Anthropic's docs: https://platform.claude.com/docs/en/build-with-claude/fast-mode

Note: As of September 1, 2026, this dedicated fast model is deprecated. Fast mode is now served by the fast service tier endpoint on Claude Opus 5 — request it with service_tier: "fast" or speed: "fast". Requests to this model keep working and are served by the same fast tier capacity, so no action is required, but new integrations should target the regular model. Learn more in our service tier docs: https://openrouter.ai/docs/guides/features/service-tiers

TexteContexte de 1M
Claude Opus 5

Claude Opus 5 is Anthropic’s flagship model for demanding reasoning, coding, and long-horizon agentic work. It is particularly strong at end-to-end software tasks, code review and bug finding, visual analysis of charts and documents, complex office deliverables, and coordinating parallel subagents.

The model maintains strong instruction following and tool use across extended tasks, while remaining effective at lower effort settings for workloads that prioritize latency and token efficiency.

TexteContexte de 1M5 $US / 25 $US
Claude Opus 5 (batch)

Claude Opus 5 is Anthropic’s flagship model for demanding reasoning, coding, and long-horizon agentic work. It is particularly strong at end-to-end software tasks, code review and bug finding, visual analysis of charts and documents, complex office deliverables, and coordinating parallel subagents.

The model maintains strong instruction following and tool use across extended tasks, while remaining effective at lower effort settings for workloads that prioritize latency and token efficiency.

TexteContexte de 1M2,50 $US / 12,50 $US
Claude Sonnet 5

Sonnet 5 is Anthropic's most capable Sonnet-class model, with frontier performance across coding, agents, and professional work. It supports adaptive thinking with selectable reasoning effort levels (low, medium, high, max, and x-high), a 1M-token context window, and text, image, and file inputs. Sonnet 5 uses an updated tokenizer and includes real-time cyber safeguards that block certain high-risk dual-use activities.

TexteContexte de 1M2 $US / 10 $US
Claude Sonnet 5 (batch)

Sonnet 5 is Anthropic's most capable Sonnet-class model, with frontier performance across coding, agents, and professional work. It supports adaptive thinking with selectable reasoning effort levels (low, medium, high, max, and x-high), a 1M-token context window, and text, image, and file inputs. Sonnet 5 uses an updated tokenizer and includes real-time cyber safeguards that block certain high-risk dual-use activities.

TexteContexte de 1M1 $US / 5 $US
Claude Fable Latest

This model always redirects to the latest model in the Claude Fable family.

TexteContexte de 1M
Claude Fable 5

Claude Fable 5 is a Mythos-class model from Anthropic, built for autonomous knowledge work and coding. It supports text, image, and file inputs with text output, with reasoning support and a 1M-token context window. It is suited for long-running, complex, and asynchronous tasks that previously required frequent human check-ins.

It is particularly strong at end-to-end work that would otherwise take a person hours, days, or weeks - taking on problems that are long-running, ambiguous, or highly multi-step. It executes well-scoped tasks with few mistakes, automatically self-correcting through verification loops, and ships with robust safeguards.

TexteContexte de 1M10 $US / 50 $US
Claude Fable 5 (batch)

Claude Fable 5 is a Mythos-class model from Anthropic, built for autonomous knowledge work and coding. It supports text, image, and file inputs with text output, with reasoning support and a 1M-token context window. It is suited for long-running, complex, and asynchronous tasks that previously required frequent human check-ins.

It is particularly strong at end-to-end work that would otherwise take a person hours, days, or weeks - taking on problems that are long-running, ambiguous, or highly multi-step. It executes well-scoped tasks with few mistakes, automatically self-correcting through verification loops, and ships with robust safeguards.

TexteContexte de 1M5 $US / 25 $US
Claude Opus 4.8

Fast-mode variant of Opus 4.8 - identical capabilities with higher output speed at 2x pricing relative to regular Opus 4.8.

Learn more in Anthropic's docs: https://platform.claude.com/docs/en/build-with-claude/fast-mode

Note: As of September 1, 2026, this dedicated fast model is deprecated. Fast mode is now served by the fast service tier endpoint on Claude Opus 4.8 — request it with service_tier: "fast" or speed: "fast". Requests to this model keep working and are served by the same fast tier capacity, so no action is required, but new integrations should target the regular model. Learn more in our service tier docs: https://openrouter.ai/docs/guides/features/service-tiers

TexteContexte de 1M
Claude Opus 4.8

Claude Opus 4.8 is Anthropic's most capable generally available model in the Opus family. It supports text, image, and file inputs with text output, with reasoning support and a 1M-token context window. It is suited for highly autonomous agents, long-horizon agentic work, knowledge work, and memory-driven tasks where coherence over extended sessions matters.

It is particularly strong on multi-step reasoning, complex coding, and end-to-end project orchestration - large codebases, multi-stage debugging, and long-running asynchronous agent pipelines. Beyond coding, it handles knowledge work such as drafting documents, building presentations, and analyzing data, maintaining quality across very long outputs.

TexteContexte de 1M5 $US / 25 $US
Claude Opus 4.8 (batch)

Claude Opus 4.8 is Anthropic's most capable generally available model in the Opus family. It supports text, image, and file inputs with text output, with reasoning support and a 1M-token context window. It is suited for highly autonomous agents, long-horizon agentic work, knowledge work, and memory-driven tasks where coherence over extended sessions matters.

It is particularly strong on multi-step reasoning, complex coding, and end-to-end project orchestration - large codebases, multi-stage debugging, and long-running asynchronous agent pipelines. Beyond coding, it handles knowledge work such as drafting documents, building presentations, and analyzing data, maintaining quality across very long outputs.

TexteContexte de 1M2,50 $US / 12,50 $US
Claude Opus 4.7

Fast-mode variant of Opus 4.7 - identical capabilities with higher output speed at premium 6x pricing.

Learn more in Anthropic's docs: https://platform.claude.com/docs/en/build-with-claude/fast-mode

TexteContexte de 1M