Meta's first frontier model — built for long-running agents, frontier coding and advanced computer use, with a 1M-token context and native multimodal input.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-wonder",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Overview
Muse Wonder is Meta's first frontier model, released October 12, 2026 by Meta Superintelligence Labs as the successor to Muse Spark 1.3. It targets personal and professional agentic work: frontier coding, long-running multi-step agents, advanced computer use, and multimodal understanding across text, image, video and audio, with a 1M-token context window.
On Venice, Muse Wonder runs under the anonymized privacy tier — your prompts are not stored, profiled, or fed back into anyone's training data. You get the same frontier agentic model Meta serves through its own API, but with zero retention and no ad-platform attachment, on a permissionless pay-per-token basis instead of a subscription gate.
Agent quickstart
The API is OpenAI-compatible: change the base URL and the model id and existing client code works unchanged.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-wonder",
"stream": true,
"messages": [{ "role": "user", "content": "Draft the release note." }]
}'curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-wonder",
"messages": [{ "role": "user", "content": "Find the rate limits." }],
"tools": [{
"type": "function",
"function": {
"name": "search_docs",
"description": "Search the API documentation.",
"parameters": {
"type": "object",
"properties": { "query": { "type": "string" } },
"required": ["query"]
}
}
}],
"tool_choice": "auto"
}'import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["VENICE_API_KEY"],
base_url="https://api.venice.ai/api/v1",
)
resp = client.chat.completions.create(
model="muse-wonder",
messages=[{"role": "user", "content": "Build without permission."}],
)
print(resp.choices[0].message.content)Specifications
Assessment
Use cases
Prompting
Hand it a goal, not a step list — describe the outcome and let it plan, delegate to sub-agents, and recover from errors on its own.
Attach screenshots or PDFs directly instead of transcribing them; vision and document parsing are native inputs.
Use function calling with a JSON schema when you need machine-readable output rather than prose.
Lean on web search in the prompt for anything time-sensitive rather than trusting training-data recall.
Reuse a cached system prompt for repeated agent runs — cached input on Venice bills at $0.25 per 1M tokens instead of $5.
For coding tasks, name the repository structure and verification steps up front; the model is tuned to make and verify changes across multiple files in parallel.
Alternatives
| Model | Best for | Context | Open weights | Price (Venice) |
|---|---|---|---|---|
| Muse Wonder | Long-running agents & frontier coding | 1M tokens | No | $5 in · $25 out / 1M |
| Kimi K3 | Open-weight agentic work | 1M tokens | Yes | $3.75 in · $18.75 out / 1M |
| GLM 5.3 Flash | Budget 1M-context work | 1M tokens | Yes | $0.15 in · $0.50 out / 1M |
| Gemini 3.8 Flash | Speed & high volume | 1M tokens | No | $0.94 in · $4.69 out / 1M |
| Qwen 3.8 Max | Frontier general reasoning | 1M tokens | No | $2.50 in · $7.50 out / 1M |
Pick Muse Wonder when the job is a long, multi-step agent run — sustained coding across a repository, computer use, or hours-long workflows where active memory matters. For the same context at a fraction of the price, GLM 5.3 Flash and Gemini 3.8 Flash cover high-volume work.
Pricing
Billed per token on Venice: $5 per 1M input tokens and $25 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Getting a key
Create a key in API settings. Nothing else is required to start.
Export it as VENICE_API_KEY so the snippets above run unedited.
Point an existing OpenAI client at https://api.venice.ai/api/v1. The scheme is part of the value: an OpenAI client given a bare host does not resolve it.
Pass muse-wonder as the model and send the request.
FAQ
Muse Wonder is Meta's first frontier model, released October 12, 2026 by Meta Superintelligence Labs. It is aimed at personal and professional agentic tasks: frontier coding, long-running agents that plan and delegate work, advanced computer use, and multimodal understanding of text, images, video, audio and PDFs, all over a 1M-token context window.
On Venice, Muse Wonder is billed per token at $5 per 1M input tokens and $25 per 1M output tokens, with cached input at $0.25 per 1M. On Meta's own API the model lists at $4/$20 per million tokens; Venice's rate differs, and there is no subscription required — you pay only for what you use.
Neither. Muse Wonder is a proprietary, closed-weights model from Meta — it cannot be self-hosted or fine-tuned, and it is not offered as a free open-source release. On Venice you pay per token, and new accounts get free welcome credits to try it. If open weights matter, Kimi K3 or GLM 5.3 Flash are the closest alternatives on Venice.
Muse Wonder is the stronger pick for long-running agentic work, computer use, and sustained coding sessions, backed by Meta's active-memory design. Kimi K3 offers open weights you can self-host, the same 1M-token context, and cheaper input pricing ($3.75 vs $5 per 1M). Choose Wonder for frontier agent depth; choose Kimi K3 for sovereignty and auditability.
Yes. Function calling is a core capability — the model is designed to call tools in parallel, delegate to sub-agents, and recover from errors during multi-step tasks. On Venice it also supports structured output via JSON schema, web search, and vision, video and audio input.
Muse Wonder has a 1,024K-token context window (about 1M tokens) with up to 128K tokens of output per response on Venice. Meta's model card confirms 1M-token context at inference, and there is no long-context price premium on the model's rate card.
Venice runs Muse Wonder under its anonymized privacy tier: prompts are not stored, profiled, or used for training, and usage is tied to no personal history. You can use it in the Venice app or through the Venice API. Note that this tier is anonymized rather than end-to-end encrypted or TEE-isolated.
Muse Wonder Full is Meta's larger, more capable sibling, scoring 56 vs 52 on the Artificial Analysis Intelligence Index and topping agentic and coding benchmarks like DeepSWE and SWE-Marathon. However, Full is exclusive to Muse Code subscribers and is not served through any API — so on Venice, Muse Wonder is the variant you get.
Muse Wonder was released on October 12, 2026, alongside Muse Wonder Full, as Meta Superintelligence Labs' first frontier model family. It has been available on Venice since October 2026.
One key, free to start, no credit card.