le robust ouvert Whisper checkpoint pour multilingue transcription et captioning.
Technical Specifications
- Provider
- OpenAI
- Category
- open-source
- Release Date
- 2023-11
- Batch API Discount
- 50% off async batch
- Consumer Subscription
- Free / Self-Hosted
- Reasoning Effort Control
- Fixed
- Structured Outputs
- No
- Output Modalities
- text
- Official Documentation
- Visit Official Docs ↗
Data Privacy, Licensing & Deployment Governance
Official Provider Release Intelligence, API Quotas & Agentic Specs
- Official API Model ID (SDK String)
- whisper-large-v3
- Supported SDK & API Protocol
- OpenAI Responses API & Chat Completions v1
- Lifecycle & Deprecation Status
- Legacy Stable (Upgrade path available in OpenAI lineup)
- Production Rate Limits (RPM & TPM)
- Unlimited (Hardware-bound on self-hosted GPU) / 10,000 RPM Cloud · Unlimited local throughput / 10M+ TPM Managed Cloud
- Long-Context Pricing Tier (>128K / >200K)
- Flat rate ($0.006/1M in) across full 0K window
- Prompt Cache Write & TTL Policy
- Automatic prefix caching at $0/1M (no extra cache-write fee)
- Computer Use & Browser GUI Agent
- Text-only completion (No GUI action loop)
- MCP & Parallel Tool Calling
- Prompt-based JSON extraction only
- Web Search Grounding & Code Sandbox
- Native Web Search tool + Hosted Code Interpreter & File Search RAG
- Fine-Tuning & Distillation Support
- Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory)
- Multimodal Tokenization Billing
- Text BPE (~4 chars/token) · Audio (~25–32 tokens/sec)
- Weights Precision & Quantization
- BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0)
- Minimum GPU VRAM Requirement
- 2 GB VRAM (4-bit Quantized) · 4 GB VRAM (Native BF16)
- Supported Inference Engines
- vLLM, SGLang, Ollama, llama.cpp, TensorRT-LLM, HuggingFace TGI
- System Card & Safety Framework
- OpenAI Preparedness Framework Audited & System Card Published
- Enterprise Compliance (SOC2 / HIPAA / ZDR)
- 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control
Whisper Large v3 Cost, Context, Hardware & API Integration
Pay-as-you-go API is cheaper for this volume (saves $19.82/mo vs $20 flat plan).
Break-even point: ~55,556 requests/day equals $20/month.
Same Monthly Workload Cost Across Popular Reference Models
Live Prompt Paste Tokenizer & Cost Counter
Paste real system instructions, JSON schemas, or code to measure BPE token counts across tokenizers.
Hybrid Prompt Routing Cascade Simulator
Route routine queries to a fast lightweight model ($0.15/1M) and escalate only complex reasoning turns to Whisper Large v3:
Benchmark Scores & 6-Axis Capability Radar
- SWE-bench Verified
- 36%
- GPQA Diamond
- 46%
- MMLU-Pro
- 66%
- HumanEval
- 82.5%
- Humanity's Last Exam
- 5.5%
- Terminal-Bench
- 23%
- SciCode
- 24.5%
- LiveCodeBench
- 39%
- AIME
- 46%
- Long-Context Reasoning
- 61%
- Chatbot Arena Elo
- 1255pts
Strengths
- Supports 99+ langues avec solide précision
- à poids ouverts pour auto-hébergement et ajustement fin
- Robust sur noisy et accented audio
- très solide sur accented et noisy parole
Limitations & Trade-offs
- pas en temps réel — batch ou chunked traitement seulement
- No conçu-dans diarisation ou speaker labels
Editorial Assessment
le standard open source ASR baseline.
Best for: Batch transcription, subtitling et multilingue captioning
Compare Whisper Large v3
Frequently Asked Questions about Whisper Large v3
How much does the Whisper Large v3 API cost per 1M tokens?+
Whisper Large v3 by OpenAI is priced at $0.006 per 1M input tokens and $0 per 1M output tokens. With prompt caching enabled, cached input tokens cost $0 per 1M tokens, and asynchronous Batch API requests receive a 50% discount.
Should I use a $20/month subscription or the Whisper Large v3 pay-as-you-go API?+
A flat $20/month consumer subscription is predictable for daily interactive chat and manual coding. However, if you are building software, running automated workflows, or sending low-to-moderate volume with prompt caching enabled, the pay-as-you-go API is typically far more cost-effective and offers higher rate limits without manual UI caps.
How many words and pages fit in Whisper Large v3's context window?+
Whisper Large v3 features a 0K tokens context window, which equals approximately 336 English words or ~1 standard single-spaced pages in a single prompt. Its maximum output generation per turn is 448 tokens.
Can I run Whisper Large v3 locally on my own GPU (Open Weights)?+
Yes, Whisper Large v3 (1.55B) releases open weights that can be self-hosted locally or on private cloud GPUs using vLLM, Ollama, SGLang, or llama.cpp.
Does OpenAI train on data sent to the Whisper Large v3 API, and is commercial use allowed?+
100% Isolated when self-hosted on private GPUs or VPC. License tier: Open-Weights Research & Commercial License. Full commercial ownership of generated outputs. Supported deployment channels include: Self-Hosted GPU (vLLM / Ollama), Private VPC, or Managed Serverless API.
How does Whisper Large v3 perform on coding and reasoning benchmarks?+
Whisper Large v3 achieves a composite Intelligence Index of 45/100, scoring 36% on SWE-bench Verified, 46% on GPQA Diamond, 66% on MMLU-Pro, and 1255 pts on Chatbot Arena Elo.
What is Whisper Large v3 best used for?+
Whisper Large v3 is best suited for: Batch transcription, subtitling et multilingue captioning. Assessment: le standard open source ASR baseline.
OpenAI Family Lineage, Routing Tiers & Upgrade Path
Step down to a lighter sibling for high-volume routing or step up to OpenAI's flagship reasoning tier:
GPT-6 Astra
GPT-6.1 Sol
GPT-6 Sol
GPT-6 Luna
GPT-5.6 Sol
GPT-5.6 Terra
AI Glossary & Architecture Concepts
View full glossary →Key machine learning concepts and benchmarks relevant to Whisper Large v3:
L'unité textuelle de base traitée par un modèle linguistique, représentant généralement 3 à 4 caractères ou environ 0,75 mot. Par exemple, une phrase courante est découpée en plusieurs jetons.
Une suite de tests standardisés permettant d'évaluer et de comparer objectivement les performances des modèles d'IA sur des tâches ciblées (code avec SWE-bench, mathématiques avec AIME, etc.).
Le nombre maximal de jetons qu'un modèle peut traiter simultanément (entrée et sortie comprises). Une fenêtre de 1 000 000 jetons permet d'analyser environ 750 000 mots en une seule requête.
L'architecture de réseau neuronal introduite en 2017 à la base des LLM modernes, utilisant l'auto-attention pour traiter tous les jetons en parallèle plutôt que séquentiellement.
Les scores de benchmarks, tarifs par token et limites évoluent rapidement. Vérifiez toujours les spécifications actuelles sur la documentation officielle du fournisseur.