Models
Updated
BroadNet Inference serves four chat models and two speech-to-text models through one OpenAI-compatible API. Filter them by input type and access, and compare up to three side by side. Gemma and both transcription models can use the starting credit; DeepSeek, GLM and Qwen3.8 need a top-up. Inference content is not retained.

Prices on this page are from BroadNet's price feed snapshot dated , not checked live here. See official pricing for current rates.
official pricingFind and compare models
6 live models.
DeepSeek V4.1 Flash
broadnet/deepseek-v4.1-flash- Text
- Vision (image input)
- Reasoning
Fast reasoning model with a 1M-token context. The only model on the unlimited plans.
- Inputs
- Text, image (vision)
- Context
- 1,048,576 tokens
- Price, USD per 1M tokens
- $0.175 input · $0.0055 cached · $0.52 output
- Access
- Top-up required
GLM 5.3
broadnet/glm-5.3- Text
- Tool calling
- Reasoning
- Structured output
Text chat with tools, structured output and reasoning, with a 1M-token context.
- Inputs
- Text
- Context
- 1,048,576 tokens
- Price, USD per 1M tokens
- $0.78 input · $0.173 cached · $3.46 output
- Access
- Top-up required
Qwen3.8 27B
broadnet/qwen3.8-27b- Text
- Vision (image input)
- Tool calling
- Reasoning
- Structured output
Text and vision chat with tools and reasoning. Tested in Arabic by BroadNet.
- Inputs
- Text, image (vision)
- Context
- 262,144 tokens
- Price, USD per 1M tokens
- $0.175 input · $0.0435 cached · $2.17 output
- Access
- Top-up required
BroadNet Gemma 4 VL 26B
broadnet/gemma-4-26b-a4b- Text
- Vision (image input)
Low-cost text and vision model. The only chat model you can use without a top-up, and the hourly-key model.
- Inputs
- Text, image (vision)
- Context
- 32,768 tokens
- Price, USD per 1M tokens
- $0.10 input · cached rate not published · $0.22 output
- Access
- Starting credit eligible
BroadNet Transcribe Arabic
broadnet/cohere-transcribe-arabic- Audio (speech to text)
- Arabic
Arabic speech to text, priced per audio minute.
- Inputs
- Audio (WAV or MP3)
- Price
- $0.0042 per audio minute
- Languages
- Arabic
- Access
- Starting credit eligible
BroadNet Transcribe
broadnet/cohere-transcribe- Audio (speech to text)
- Arabic
Speech to text in 14 languages, priced per audio minute.
- Inputs
- Audio (WAV or MP3)
- Price
- $0.002 per audio minute
- Languages
- 14 languages
- Access
- Starting credit eligible
Compare selected models
Rates from the price feed snapshot; capabilities as BroadNet's API reference documents them. "Not published" means BroadNet does not state it, not that it is zero or unsupported.
Next step. The BroadNet console does not receive this comparison. Copy the ID of the model you choose from the table, then create a key and send that ID as model.
All published rates
USD per 1M tokens for chat, per audio minute for speech to text. Access comes from the price feed's paid_only field. Estimate a monthly charge or see the DeepSeek session plans.
| Chat model | Input, per 1M tokens | Cached input, per 1M tokens | Output, per 1M tokens | Context, tokens | Input types | Access |
|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash broadnet/deepseek-v4.1-flash | $0.175 | $0.0055 | $0.52 | 1,048,576 | text, image (vision) | Top-up required |
| GLM 5.3 broadnet/glm-5.3 | $0.78 | $0.173 | $3.46 | 1,048,576 | text | Top-up required |
| Qwen3.8 27B broadnet/qwen3.8-27b | $0.175 | $0.0435 | $2.17 | 262,144 | text, image (vision) | Top-up required |
| BroadNet Gemma 4 VL 26B broadnet/gemma-4-26b-a4b | $0.10 | Not published | $0.22 | 32,768 | text, image (vision) | Starting credit eligible |
| Speech to text | USD per audio minute | Languages | Access |
|---|---|---|---|
| BroadNet Transcribe Arabic broadnet/cohere-transcribe-arabic | $0.0042 | Arabic | Starting credit eligible |
| BroadNet Transcribe broadnet/cohere-transcribe | $0.002 | 14 languages | Starting credit eligible |
Which model should I use?

- Coding agents and long documents: DeepSeek V4.1 Flash. A 1M-token context, and the only model you can run unlimited on a plan.
- Tool calling and structured output: GLM 5.3, also with a 1M-token context.
- Arabic text and images: Qwen3.8 27B, which BroadNet tested in Arabic, including Gulf and Egyptian dialects.
- Lowest BroadNet per-token price, or trying the API on the starting credit: BroadNet Gemma 4 VL 26B.
- Arabic voice notes and calls: BroadNet Transcribe Arabic.
- Speech in other languages: BroadNet Transcribe, 14 languages.
Which models are available on request?

BroadNet lists these as available on request. They are not on the public rate card, so they have no published price. BroadNet says it issues each customer a private model ID on the same API. Ask BroadNet before relying on any of them. What BroadNet publishes about each:

- Cohere Transcribe Arabic, BroadNet fine-tune: tuned on BroadNet's own call audio for phone numbers, amounts, dates, order codes and names in narrowband and noisy audio. It can be tuned further on your recordings and vocabulary.
- Qwen3-VL 30B-A3B Instruct: a mixture of experts with 3B active parameters, 4-bit AWQ, 16K context, text and image in. BroadNet measures 36.5 to 110 tokens per second per request.
- Qwen3-VL 8B Instruct: dense 8B, 8-bit AWQ, 32K context. BroadNet reports 80.24 on its OmniDocBench subset and 73.4 on Arabic MDPBench, the best Arabic document reader it has measured.
Private endpoints and additional models are arranged by enquiry, not self-service. Ask BroadNet about a private endpoint.
Where do the models run?
BroadNet's privacy statement says its own models and the open-weight models it hosts run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup, and that partner-served models run under zero-data-retention terms with US inference partners. Its Terms of Service name DeepSeek V4.1 Flash as partner-hosted. BroadNet's model pages say the two transcription models run on NVIDIA GPUs it operates in the United States, and Gemma 4 VL 26B on NVIDIA Blackwell-class GPUs with a hot backup. BroadNet does not say on its pages where GLM 5.3 and Qwen3.8 27B are served for API traffic. For data-residency needs, BroadNet asks you to agree processing locations before you send data; its inference page says regional processing in MENA or Europe is available on request.
The hardware behind the API, as BroadNet describes it
BroadNet reports running about 270 models from about 40 families and vendors on its own NVIDIA and AMD GPUs, across 13 GPU families in 3 tiers, from workstation RTX cards to data-center B300-class accelerators, measured in its lab in September 2026. It publishes the models it rules out, with reasons, alongside the ones it ships.
- BroadNet's GPU hardware and serving approach: tiers, throughput per GPU and what BroadNet says its hardware taught it.
- Models BroadNet evaluated and ruled out, and why.
- Try the models in your browser before you create an account.