Orpheus 3B
Orpheus 3B is a 3-billion-parameter English text-to-speech model from Canopy Labs, optimized for natural prosody, expressive delivery, and real-time streaming speech generation.
What is Orpheus 3B?
Orpheus 3B is a 3B-parameter Llama-based text-to-speech model by Canopy Labs that converts text into natural-sounding, emotionally expressive speech. It is primarily used for applications like narration, audiobooks, and content creation where high-quality, human-like delivery is important. It is also used in voice assistants, interactive agents, and real-time conversational interfaces that require low-latency streaming audio and multiple voice options. Orpheus 3B belongs to the Orpheus TTS family of Llama-based speech-LLMs, which includes multilingual and fine-tuned variants building on the orpheus-3b-0.1-pretrained base model.
Providers
Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick — Balanced (price + speed), Nitro (fastest), or Exacto (highest tool-calling accuracy).
| Provider | Input | Output | Cache read /M | Latency | Throughput | Uptime |
|---|---|---|---|---|---|---|
| Canopy Labs | ~$0.60 | ~$0.60 | — | ~120ms | ~140 tps | ~99.9% |
| OpenAI | ~$0.80 | ~$0.80 | — | ~110ms | ~160 tps | 99.06% |
| Anthropic | ~$0.90 | ~$0.90 | — | ~130ms | ~120 tps | 99.28% |
| Google Cloud | ~$0.70 | ~$0.70 | — | ~140ms | ~150 tps | ~99.9% |
Try this model
Test Orpheus 3B right here — free to start.
Suggestions for your first prompt
Code snippet
Call the model through the OpenAI-compatible API.
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://inference.example.com/v1"
)
response = client.chat.completions.create(
model="canopy-labs/orpheus-3b",
messages=[
{
"role": "user",
"content": "Describe this image in one sentence."
}
],
)
print(response.to_json())
{
"model": "canopy-labs/orpheus-3b",
"messages": [
{
"role": "user",
"content": "Describe this image in one sentence."
}
]
}
5 Core Capabilities
-
Natural TTS
Generates high-quality, natural-sounding English speech with expressive prosody, suitable for narration, assistants, and interactive applications.
-
Voice Variety
Provides multiple preset voices with distinct characteristics, enabling flexible voice selection for different products, brands, and use cases.
-
Emotion Control
Supports guided emotion and intonation via simple tags, allowing control over expressiveness like laughter, sighs, or other vocal nuances.
-
Zero-Shot Cloning
Enables zero-shot voice cloning from short audio samples, producing personalized synthetic voices without task-specific fine-tuning.
-
Low-Latency Streaming
Optimized for low-latency streaming inference, delivering near real-time audio suitable for interactive conversational and live applications.
6 Most Valuable Use Cases
- Audiobook Narration
- Voice Assistants
- Interactive Storytelling
- Podcast Voice Generation
- Zero-Shot Voice Cloning
- Emotion-Rich Dialogues
Why Build on LLM.API?
One unified API. Every major model. Built-in reliability, cost control, and observability.
-
Intelligent Model Routing
Dynamically route each request to the best-fit model across providers based on task, latency, and reliability—without changing your integration.
One endpoint, any model -
Cost-Aware Orchestration
Automatically optimize for price-performance by mixing premium and budget models, enforcing per-request budgets, and tracking spend in one place.
More output per dollar -
Resilient Fallback Flows
Define multi-provider fallback chains so requests transparently retry on alternate models when a vendor is slow, degraded, or down.
Stay online, automatically -
Full-Stack Observability
Get unified logs, traces, and metrics across all providers with per-model timing, errors, and payloads for fast debugging and tuning.
See every token hop -
Task-Level Abstractions
Call high-level task APIs (chat, tools, RAG, vision, structured outputs) instead of provider-specific endpoints, and swap models without rewriting logic.
Code to tasks, not vendors -
High-Throughput Batch Jobs
Run large batch inference workloads with automatic chunking, retries, and aggregation, maximizing throughput while staying within provider limits.
Ship millions of calls
When to Use — When NOT to Use
Use it if...
- You need a lightweight general-purpose model for simple chatbots and FAQ assistants.
- You need cost-efficient inference for high-volume, low-complexity classification or tagging tasks.
- You need to embed the model on modest cloud instances with limited GPU memory.
- Your use case involves short-form content generation like titles, snippets, and brief summaries.
- Your use case involves rapid experimentation where fast cold-start and iteration speed matter.
- Your use case involves basic structured output generation such as JSON objects or short forms.
- You need a model small enough for on-demand autoscaling without high infrastructure cost.
Avoid if...
- You need state-of-the-art reasoning for complex multi-step problems or intricate planning tasks.
- Your workload requires handling very long contexts like full books or extensive logs.
- You need highly creative, long-form writing comparable to top-tier frontier language models.
- You need robust performance on advanced math, formal proofs, or symbolic reasoning workloads.
- Your workload requires best-in-class code generation, debugging, and multi-file repository understanding.
- You need strong multilingual support across many low-resource languages with high accuracy.
- Your workload requires strict enterprise-grade safety, red-teaming, and domain-specific guardrails out-of-box.
Frequently Asked Questions
-
What is Orpheus 3B?
Orpheus 3B is a 3-billion-parameter language model from Canopy Labs optimized for fast, low-cost text generation and code completion via LLM.API.
-
What is Orpheus 3B best suited for?
Orpheus 3B is best for lightweight chatbots, code assistants, and high-volume text generation where low latency and inexpensive inference are critical.
-
What is the context window of Orpheus 3B?
Orpheus 3B supports a context window of up to 8,192 tokens per request via LLM.API.
-
How much does it cost to use Orpheus 3B on LLM.API?
Orpheus 3B pricing on LLM.API is per-token; check the LLM.API pricing page for the latest input and output token rates.
-
How fast is Orpheus 3B in terms of latency and throughput?
Orpheus 3B is designed for low p95 latency on short prompts and supports high request throughput suitable for production workloads.
-
What modalities does Orpheus 3B support?
Orpheus 3B is a text-only model that accepts text prompts and returns text completions.
-
How do I call Orpheus 3B through the LLM.API gateway?
Select Orpheus 3B as the model in your LLM.API completion or chat endpoint request and authenticate using your LLM.API key.
-
How does Orpheus 3B compare to larger models available on LLM.API?
Compared to larger models, Orpheus 3B is cheaper and faster but generally less capable on complex reasoning and long-context tasks.
-
Does Orpheus 3B support function calling or tool usage?
Orpheus 3B can be wrapped in LLM.API’s tool-calling interfaces, but complex tool orchestration may work better with larger models.
-
What are the main limitations of Orpheus 3B?
Orpheus 3B may struggle with nuanced reasoning, very long documents, domain-expert knowledge, and strict factual accuracy compared to larger frontier models.
COMPARE
Competitive Models
-
Gemini 3.1 Flash TTS Preview
Gemini 3.1 Flash TTS Preview is Google’s low-latency text‑to‑speech model that generates natural, expressive speech with fine-grained control via style prompts and audio tags. It is optimized for fast, high‑quality voice synthesis across many languages and voices.
-
GLM 4.7
GLM 4.7 is Z.ai’s flagship large language model, optimized for strong coding performance and stable multi-step reasoning. It is notable for its very large context window and open-source availability under Apache 2.0.
-
Qwen3.5-9B
Qwen3.5-9B is a 9‑billion‑parameter multimodal language model from Qwen that supports long-context reasoning over text and images. It is designed to offer strong reasoning, coding, and visual understanding capabilities in a relatively compact, efficient architecture.
Get one key to every model
Swap your API key. Keep your code.