whisper-medium.en
Whisper is a pre-trained model for automatic speech recognition (ASR) and speech translation. Trained on 680k hours of labelled data, Whisper models demonstrate a strong ability to generalise to many datasets and domains without the need for fine-tuning.
Fact sheet
Parameters
Context length
Architecture
Shape
Weights stored as
Task
Languages
Built on
Released
Last updated
Downloads (last 30 days)
Licence and access
Licence
Access
Security
Hugging Face security scan
Weight files by format
Facthory’s checks
Safetensors or GGUF files beside pickle-based ones.
No file flagged.
The card declares its licence.
The card reports safety or harm evaluations with their results.
GPU calculator
How much memory this model needs to serve at a quantization, context and number of concurrent sequences, which GPUs hold it, and how fast it can generate. Estimates from the model’s published size and shape, with the assumptions below.
Weights
Context per sequence: 8K tokens
Concurrent sequences: 1
KV cache
Memory
1.7 GB
Weights: 1.5 GB
KV cache: –
Runtime: 0.2 GB
The model’s shape is not published, so the KV cache is not included.
| GPU | Memory | GPUs needed | Per sequence | All sequences |
|---|---|---|---|---|
| NVIDIA L4 | 24 GB | 1 × | 118 tokens/s | 118 tokens/s |
| NVIDIA L40S | 48 GB | 1 × | 339 tokens/s | 339 tokens/s |
| NVIDIA RTX 6000 Ada | 48 GB | 1 × | 377 tokens/s | 377 tokens/s |
| NVIDIA RTX PRO 6000 Blackwell | 96 GB | 1 × | 704 tokens/s | 704 tokens/s |
| NVIDIA A100 80GB SXM | 80 GB | 1 × | 801 tokens/s | 801 tokens/s |
| NVIDIA H100 SXM | 80 GB | 1 × | 1,316 tokens/s | 1,316 tokens/s |
| NVIDIA H200 SXM | 141 GB | 1 × | 1,885 tokens/s | 1,885 tokens/s |
| AMD Instinct MI300X | 192 GB | 1 × | 2,082 tokens/s | 2,082 tokens/s |
| AMD Instinct MI325X | 256 GB | 1 × | 2,356 tokens/s | 2,356 tokens/s |
Assumptions
- Runtime memory (activations, buffers) is 10% of weights and KV cache, plus 1 GB per GPU for the framework.
- A GPU is filled to at most 90% of its memory; tensor parallelism uses 1, 2, 4 or 8 GPUs in one server.
- Generation is limited by memory bandwidth at 60% of the vendor’s peak: each step reads the weights once and every sequence’s KV cache. A mixture-of-experts model reads only its active experts, so it runs faster than shown.
- GPU memory and bandwidth are the vendors’ published figures.
Profile
Kind
Lineage
Family
Uses
In and out
Written on 11 Nov 2026 by Workers AI (@cf/zai-org/glm-4.7-flash) from the model card and the hub’s metadata: a reading of the publisher’s text, not checked by Facthory.
Published benchmarks
As the model card publishes them, with the source the card names. Facthory has not run them; results are self-reported unless the hub verified them.
| Benchmark | Metric | Result | Source | Counts toward |
|---|---|---|---|---|
| LibriSpeech (clean) clean · test Not counted: not on our list | Test WER | 4.12 | Model card | Not counted: not on our list |
| LibriSpeech (other) other · test Not counted: not on our list | Test WER | 7.43 | Model card | Not counted: not on our list |
The capability score counts results the model’s authors publish, for the benchmarks our methodology lists, where at least two models in the catalog report the same benchmark, configuration and metric. Results a card attributes to the Open LLM Leaderboard are the leaderboard’s runs, and Epoch AI’s evaluations are Epoch’s: neither counts. How we score capability
EU AI Act
What the EU AI Act asks of a general-purpose AI model depends on who places it on the EU market and how. These facts bear on it:
- Licence: apache-2.0
- Access: Weights download without approval
- Parameters: 763.9M
General-purpose AI model (Article 3(63))
Not indicated. Its task (automatic-speech-recognition) generates no text, audio, images or video, which the Commission’s indicative criterion asks for.
Articles 51 to 55, the obligations for general-purpose AI models, do not bind its provider.
Facthory’s EU AI Act score appears once a person has reviewed these checks. The articles we score against
