Zum Inhalt springen
Alpha
  • Modelle
  • Ranglisten
  • Ori
Vertrieb kontaktieren

Fußzeile

OpenRouterOpenRouter
© 2026 OpenRouter, Inc

Produkt

  • Chat
  • Ranglisten
  • Leistungsvergleiche
  • Apps
  • Entdecken
  • Modelle
  • Ori
  • Sammlungen
  • Anbieter
  • Tools
  • Preise
  • Business
  • Enterprise
  • Labs

Unternehmen

  • Über
  • Blog
  • Kunden
  • Karriere
    Offene Stellen
  • Datenschutz
  • Nutzungsbedingungen
  • Vertrauenszentrum
  • Support
  • Funktioniert mit OR
  • Daten
  • Marke

Entwickler

  • Dokumentation
  • API-Referenz
  • Entwicklerplattform
  • Status
  • KI-Sitemap

Verbinden

  • Discord
  • GitHub
  • LinkedIn
  • X
  • YouTube
Favicon für openai

OpenAI: GPT-5.4

openai/gpt-5.4

Vergleich

GPT-5.4 is OpenAI’s latest frontier model, unifying the Codex and GPT lines into a single system. It features a 1M+ token context window (922K input, 128K output) with support for text and image inputs, enabling high-context reasoning, coding, and multimodal analysis within the same workflow.

The model delivers improved performance in coding, document understanding, tool use, and instruction following. It is designed as a strong default for both general-purpose tasks and software engineering, capable of generating production-quality code, synthesizing information across multiple sources, and executing complex multi-step workflows with fewer iterations and greater token efficiency.

Modalitäten
Eingabe- / Ausgabepreis
2,50 $ / 15 $pro 1 Mio.
Kontext
1,1M
Veröffentlicht
5. März 2026
Vergleich
AnbieterPreiseLeistungVerfügbarkeitLeistungsvergleicheAppsAktivitätDokumentationFAQEntdecken

Anbieter

Verschiedene Unternehmen hosten dasselbe Modell. OpenRouter leitet Ihre Anfrage je nach gewähltem Routing-Modus an eines davon weiter — Balanced (Preis + Geschwindigkeit), Nitro (am schnellsten) oder Floor (am günstigsten).

Preise

Der durchschnittliche Preis, den Kunden tatsächlich für dieses Modell zahlen, neben den von Anbietern veröffentlichten Preisen. Durch Caching und Rabatte liegt der tatsächlich gezahlte Preis oft deutlich unter dem Listenpreis.

Leistung

Der Durchsatz gibt an, wie schnell das Modell schreibt (Token pro Sekunde – höher ist besser). Die Latenz ist die gesamte Umlaufzeit (niedriger ist besser). TTFT ist die Zeit bis zum ersten Token – also die Wartezeit, bis etwas angezeigt wird (niedriger ist besser).

Verfügbarkeit

Verfügbarkeit ist der prozentuale Anteil der letzten 3 Tage, in dem mindestens ein Anbieter auf Anfragen reagiert hat. Die Verfügbarkeit der Inferenz ist der prozentuale Anteil der Zeit, in der Inferenz erfolgreich bereitgestellt wurde. OpenRouter überwacht dies kontinuierlich und verwendet den nächstbesten Anbieter, wenn ein Anbieter einen Fehler zurückgibt.

Leistungsvergleiche

Ergebnisse standardisierter Bewertungen und eigene Ausgaben dieses Modells zu gemeinsamen Vergleichstest-Prompts. Höhere Prozentwerte sind besser – und jeder Rang zeigt, wo dieses Modell unter den anhand dieser Quelle bewerteten Modellen liegt.

Zusammenfassung der Benchmark-Ergebnisse für OpenAI: GPT-5.4 (Artificial Analysis, Design Arena und Vals AI)
QuelleBenchmarkErgebnis
Artificial AnalysisGPT-5.4 (Non-reasoning) GPQA Diamond74.8%
Artificial AnalysisGPT-5.4 (Non-reasoning) HLE11.3%
Artificial AnalysisGPT-5.4 (Non-reasoning) IFBench48.4%
Artificial AnalysisGPT-5.4 (Non-reasoning) τ²-Bench Telecom36.0%
Artificial AnalysisGPT-5.4 (Non-reasoning) AA-LCR58.3%
Artificial AnalysisGPT-5.4 (Non-reasoning) CritPt0.6%
Artificial AnalysisGPT-5.4 (Non-reasoning) Terminal-Bench Hard37.9%
Artificial AnalysisGPT-5.4 (Xhigh) Coding Index71,1
Artificial AnalysisGPT-5.4 (Xhigh) GPQA Diamond92.0%
Artificial AnalysisGPT-5.4 (Xhigh) HLE43.7%
Artificial AnalysisGPT-5.4 (Xhigh) IFBench73.9%
Artificial AnalysisGPT-5.4 (Xhigh) τ²-Bench Telecom87.1%
Artificial AnalysisGPT-5.4 (Xhigh) AA-LCR82.0%
Artificial AnalysisGPT-5.4 (Xhigh) τ-Bench Banking39.6%
Artificial AnalysisGPT-5.4 (Xhigh) GDPval-AA37.8%
Artificial AnalysisGPT-5.4 (Xhigh) CritPt23.4%
Artificial AnalysisGPT-5.4 (Xhigh) Terminal-Bench Hard57.6%
Artificial AnalysisGPT-5.4 (Xhigh) Terminal-Bench 2.178.3%
Artificial AnalysisGPT-5.4 (Low) GPQA Diamond87.1%
Artificial AnalysisGPT-5.4 (Low) HLE30.8%
Artificial AnalysisGPT-5.4 (Low) IFBench65.9%
Artificial AnalysisGPT-5.4 (Low) τ²-Bench Telecom74.6%
Artificial AnalysisGPT-5.4 (Low) AA-LCR76.7%
Artificial AnalysisGPT-5.4 (Low) CritPt7.4%
Artificial AnalysisGPT-5.4 (Low) Terminal-Bench Hard43.2%
Design ArenaGPT-5.4 (Medium) Models Arena 3D Elo1099
Design ArenaGPT-5.4 (Medium) Models Arena Asciiart Elo1196
Design ArenaGPT-5.4 (Medium) Models Arena Code Categories Elo1208
Design ArenaGPT-5.4 (Medium) Models Arena Data Visualization Elo1235
Design ArenaGPT-5.4 (Medium) Models Arena Game Development Elo1237
Design ArenaGPT-5.4 (Medium) Models Arena SVG Elo1191
Design ArenaGPT-5.4 (Medium) Models Arena UI Component Elo1234
Design ArenaGPT-5.4 (Medium) Models Arena Website Elo1215
Vals AIGPT 5.4 (xhigh) AIME96.7%
Vals AIGPT 5.4 (xhigh) CaseLaw v263.8%
Vals AIGPT 5.4 (xhigh) Code Migration35.0%
Vals AIGPT 5.4 (xhigh) CorpFin v265.3%
Vals AIGPT 5.4 (xhigh) Finance Agent v1.157.2%
Vals AIGPT 5.4 (xhigh) GPQA Diamond91.7%
Vals AIGPT 5.4 (xhigh) Harvey's Legal Agent Benchmark v10.0%
Vals AIGPT 5.4 (xhigh) IOI v167.8%
Vals AIGPT 5.4 (xhigh) LiveCodeBench84.1%
Vals AIGPT 5.4 (xhigh) LegalBench86.0%
Vals AIGPT 5.4 (xhigh) MedCode41.3%
Vals AIGPT 5.4 (xhigh) MedQA96.1%
Vals AIGPT 5.4 (xhigh) MedScribe77.5%
Vals AIGPT 5.4 (xhigh) MMLU Pro87.5%
Vals AIGPT 5.4 (xhigh) MMMU Pro87.5%
Vals AIGPT 5.4 (xhigh) MortgageTax68.3%
Vals AIGPT 5.4 (xhigh) ProgramBench0.0%
Vals AIGPT 5.4 (xhigh) ProofBench v156.0%
Vals AIGPT 5.4 (xhigh) SAGE v143.3%
Vals AIGPT 5.4 (xhigh) SkillsBench51.7%
Vals AIGPT 5.4 (xhigh) SWE-bench78.2%
Vals AIGPT 5.4 (xhigh) TaxEval v274.0%
Vals AIGPT 5.4 (xhigh) Terminal-Bench 2.058.4%
Vals AIGPT 5.4 (xhigh) Vals Index66.1%
Vals AIGPT 5.4 (xhigh) Vals Multimodal Index65.2%
Vals AIGPT 5.4 (xhigh) Vibe Code Bench v1.167.4%

Apps

Öffentliche Apps, die den meisten Datenverkehr an dieses Modell senden. Ein guter Anhaltspunkt dafür, wie echte Produktionsworkloads aussehen – und ein Hinweis darauf, für welche Anwendungsfälle dieses Modell am besten geeignet ist.

Aktivität

Token-Volumen und Anfrageaufkommen für dieses Modell im Zeitverlauf.

API

Direkt einsetzbarer Code zum Aufrufen dieses Modells. Die API von OpenRouter ist OpenAI-kompatibel – bei den meisten SDKs genügt es, die Basis-URL auszutauschen. Zwischen den Modellen ändert sich nur der unten stehende Modell-Slug.

Über regionales Routing innerhalb der EU verfügbar

OpenAI: GPT-5.4 ist über regionsinternes EU-Routing verfügbar. An eu.openrouter.ai gesendete Anfragen werden in der EU entschlüsselt und ausschließlich an gemeinsam genutzte Anbieter-Endpunkte in Europa (EU/EWR, Vereinigtes Königreich und Schweiz) weitergeleitet.

Anbieter in Europa
Azure
Preise in Europa
  • Azure: 2,75 $/M tokens Eingabe, 16,50 $/M tokens Ausgabe
Basis-URL
https://eu.openrouter.ai/api/v1
Tarif
Erfordert Business oder Enterprise. Self-Service mit Business: kein Vertrag, kein monatlicher Mindestbetrag. OpenRouter Business für Routing innerhalb der EU

Im Fehlerfall gesperrt: Wenn kein gemeinsam genutzter Anbieter-Endpunkt in Europa verfügbar ist, schlägt die Anfrage fehl, statt die Region zu verlassen. Anfragen, die mit Ihrem eigenen BYOK-Schlüssel bedient werden, gehen an die Region Ihrer Bereitstellung, die Sie selbst kontrollieren. Dokumentation zum regionsinternen Routing

Dokumentation

Anleitungen, Rezeptsammlungen, Blogbeiträge und Tools für die Entwicklung mit OpenAI: GPT-5.4 auf OpenRouter.

  • GPT-5.4 Migration GuideRezeptsammlung

Weitere Modelle entdecken

KI-Modelle mit Bildverarbeitung: Multimodale LLMs für das BildverständnisSammlungKI-ModellranglistenRangliste
Latenz / Durchsatz
2,50 $15,00 $0,250 $1,48s71 tps
100,00 %
2,50 $15,00 $0,250 $1,77s54 tps
99,93 %
Nicht beim Routing im Modus Standard verwendet:Warum diese Endpunkte nicht verwendet werden
2,75 $16,50 $0,275 $1,34s62 tps
100,00 %
2,75 $16,50 $0,275 $0,64s78 tps
100,00 %
2,75 $16,50 $0,275 $1,56s90 tps
--
Fast
5,00 $30,00 $0,500 $1,70s42 tps
100,00 %
Flex
1,25 $7,50 $0,125 $9,58s35 tps
48,08 %
Durchsatz
90Token/s

P50, bester Wert über alle Anbieter hinweg

Latenz
0,64s

P50, bester Anbieter

AutoExacto-Leistungsvergleiche
TAU-BenchOpenAI74,7 %
Verfügbarkeit (3 Tage)Das Modell war erreichbar. Die Anfrage wurde an einen Anbieter weitergeleitet.
100,00 %
Verfügbarkeit (3 Tage)Das Modell hat über einen beliebigen Anbieter ein Inferenzergebnis zurückgegeben. Fehler und leere Antworten wirken sich negativ darauf aus.
99,79 %

Verfügbarkeit in den letzten 3 Tagen

Letzte 72 Stunden
Verfügbarkeit 99,79 %
Vor 3 TagenVor 2 TagenGesternJetzt

Verfügbarkeit in den letzten 24 Stunden

OpenRouter Availability
99,73 %
Without Routing
97,88 %

Wenn bei einem vorgelagerten Anbieter ein Fehler auftritt, können wir diesen durch die Weiterleitung an einen anderen funktionsfähigen Anbieter beheben, sofern Ihre Anfragefilter dies zulassen. Sie können programmgesteuert auf die Verfügbarkeitsdaten pro Anbieter zugreifen, und zwar über die Endpunkt-API. Erfahren Sie mehr über unsere Optionen für Lastverteilung und Anbieter-Routing.

1.
Favicon für https://crm.sotaservers.cloud/
Lumea CRM
neu
10,9BToken
2.
Favicon für 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.
9,42BToken
3.
Favicon für https://openai.com/codex
Codex
A coding agent that helps you build and ship with AI
6,95BToken
4.
Favicon für https://www.tradingview.com/
Trading View
neu
4,98BToken
5.
Favicon für 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.
3,28BToken

Häufig gestellte Fragen

GPT-5.4 is OpenAI’s latest frontier model, unifying the Codex and GPT lines into a single system. It features a 1M+ token context window (922K input, 128K output) with support for text and image inputs, enabling high-context reasoning, coding, and multimodal analysis within the same workflow.

Die Dokumentation zu GPT-5.4 auf OpenRouter umfasst das Cookbook GPT-5.4 Migration Guide. Der Schnellstart behandelt API-Schlüssel und Ihre erste Anfrage, und die API-Referenz listet alle Parameter auf.

GPT-5.4 kostet 2,50 $/M Eingabe-Token und 15,00 $/M Ausgabe-Token, mit separaten Tarifen für Cache Read für 0,25 $/M tokens.

GPT-5.4 hat ein Kontextfenster von 1.050.000 Token. Es unterstützt bis zu 128.000 Completion-Token.

Ja. GPT-5.4 akzeptiert tools und tool_choice für Funktionsaufrufe. Es unterstützt auch strukturierte Ausgaben über ein JSON-Schema in response_format.

GPT-5.4 akzeptiert Text, Bilder und Dateien wie PDFs als Eingabe und gibt Text zurück.

GPT-5.4 wird auf OpenRouter von 3 Anbietern bereitgestellt: OpenAI, Azure und Amazon Bedrock. Anfragen werden an den besten verfügbaren Anbieter weitergeleitet, bei Ausfällen wird automatisch zu den anderen gewechselt, und Sie können Anbieter über das Anbieter-Routing festlegen oder ausschließen.

GPT-5.4 wurde am 5. März 2026 veröffentlicht.

Weitere Modelle von OpenAI

GPT-6.1 Terra Pro

GPT-6.1 Terra Pro is the same underlying model as GPT-6.1 Terra, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Terra and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Terra instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext1 $ / 5 $
GPT-6.1 Terra Pro (batch)

GPT-6.1 Terra Pro is the same underlying model as GPT-6.1 Terra, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Terra and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Terra instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext0,50 $ / 2,50 $
GPT-6.1 Terra

GPT-6.1 Terra is the mid-tier model in OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It is suited for everyday agentic, coding, and writing work, is particularly strong at understanding screenshots and on-screen content, and is efficient enough for high-volume workloads.

Text1,1M Kontext1 $ / 5 $
GPT-6.1 Terra (batch)

GPT-6.1 Terra is the mid-tier model in OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It is suited for everyday agentic, coding, and writing work, is particularly strong at understanding screenshots and on-screen content, and is efficient enough for high-volume workloads.

Text1,1M Kontext0,50 $ / 2,50 $
GPT-6 Luna Decisions

GPT-6 Luna Decisions is GPT-6 Luna served through OpenAI's Decisions API. Instead of generating text, it reads the content passed as state (text, JSON, or images) and returns typed, probabilistic answers to 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), 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. A request can carry up to 200 questions about the same content.

Entscheidungen0,10 $ / 0 $
GPT-6.1 Sol Pro

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext2 $ / 10 $
GPT-6.1 Sol Pro (batch)

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext1 $ / 5 $
GPT-6.1 Sol

GPT-6.1 Sol is an upgrade to GPT-6 Sol from OpenAI, positioned below the flagship GPT-6 Astra in the GPT-6 series. It is suited for agentic coding, computer use, document-heavy professional work, and multi-step business workflow automation, and approaches Astra-level results on these tasks at a much lower cost. Compared with GPT-6 Sol, it makes fewer factual errors and is more reliable at respecting explicit restrictions and user intent during agentic tasks.

Text1,1M Kontext2 $ / 10 $
GPT-6.1 Sol (batch)

GPT-6.1 Sol is an upgrade to GPT-6 Sol from OpenAI, positioned below the flagship GPT-6 Astra in the GPT-6 series. It is suited for agentic coding, computer use, document-heavy professional work, and multi-step business workflow automation, and approaches Astra-level results on these tasks at a much lower cost. Compared with GPT-6 Sol, it makes fewer factual errors and is more reliable at respecting explicit restrictions and user intent during agentic tasks.

Text1,1M Kontext1 $ / 5 $
GPT-6 Luna Pro

GPT-6 Luna Pro is the same underlying model as GPT-6 Luna, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Luna and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Luna instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext0,10 $ / 0,50 $
GPT-6 Luna Pro (batch)

GPT-6 Luna Pro is the same underlying model as GPT-6 Luna, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Luna and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Luna instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext0,05 $ / 0,25 $
GPT-6 Luna

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic tasks, and at higher reasoning effort it can take on complex software engineering and computer-use tasks that previously called for a Sol-tier model. It shares the GPT-6 family's gains in factual reliability and its clearer, more concise communication style.

Text1,1M Kontext0,10 $ / 0,50 $
GPT-6 Luna (batch)

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic tasks, and at higher reasoning effort it can take on complex software engineering and computer-use tasks that previously called for a Sol-tier model. It shares the GPT-6 family's gains in factual reliability and its clearer, more concise communication style.

Text1,1M Kontext0,05 $ / 0,25 $
GPT-6 Sol Pro

GPT-6 Sol Pro is the same underlying model as GPT-6 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext2 $ / 10 $
GPT-6 Sol Pro (batch)

GPT-6 Sol Pro is the same underlying model as GPT-6 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

Text1,1M Kontext1 $ / 5 $
GPT-6 Sol

GPT-6 Sol is the cost-efficient high-end model in OpenAI's GPT-6 series, positioned below the flagship GPT-6 Astra and above the fast GPT-6 Luna tier. It is suited for demanding professional work, agentic coding, business workflow automation, and computer use, and is particularly strong at long-horizon software engineering tasks in real codebases. It approaches Astra-level factual reliability at a much lower cost and shares Astra's clearer, more concise communication style in technical and coding conversations.

Text1,1M Kontext2 $ / 10 $
GPT-6 Sol (batch)

GPT-6 Sol is the cost-efficient high-end model in OpenAI's GPT-6 series, positioned below the flagship GPT-6 Astra and above the fast GPT-6 Luna tier. It is suited for demanding professional work, agentic coding, business workflow automation, and computer use, and is particularly strong at long-horizon software engineering tasks in real codebases. It approaches Astra-level factual reliability at a much lower cost and shares Astra's clearer, more concise communication style in technical and coding conversations.

Text1,1M Kontext1 $ / 5 $
GPT Astra Latest

This model always redirects to the latest model in the GPT Astra family.

Text1,1M Kontext
GPT Sol Latest

This model always redirects to the latest model in the GPT Sol family.

Text1,1M Kontext
GPT Terra Latest

This model always redirects to the latest model in the GPT Terra family.

Text1,1M Kontext