Muse Spark 1.2 is Meta's reasoning model for complex agentic tasks, and the current checkpoint of the Muse Spark family. Meta describes it as an updated checkpoint over Muse Spark 1.1 with slightly higher performance, served on the same Standard tier at identical pricing — a drop-in upgrade rather than a new tier. It accepts an unusually broad input surface — text, images, video, audio, and PDF documents — and returns text, reasoning natively across modalities within a 1M-token context window. It supports configurable reasoning effort, native tool calling, and structured outputs, making it well suited for multimodal agents, deep research over mixed media, and long-context analysis. Muse Spark 1.2 targets workloads that combine documents, screenshots, recordings, and video with text — from analyzing long reports and media libraries to driving multi-step agentic pipelines that must reason over more than just text. Meta also ships muse-spark-1.2-contributor, the same checkpoint on its Contributor tier.
Muse Spark 1.2 is a Meta-produced language model served through OrcaRouter. It accepts text, image, video, file, and audio inputs, and produces text outputs. The model has a context window of…
Based on the provided facts, Muse Spark 1.2 can take in text, image, video, file, and audio, and respond with text. It is therefore suitable for multimodal understanding tasks such as summarization, extraction, translation of audio content, and answering questions about visual or audio material. A 1,048,576-token context window means it can use substantial source material as context. The model weighs all input modalities against the same token budget. There is no information in the supplied metadata about special capabilities such as tool calling, structured output, or fine-tuning. OrcaRouter offers the model through a standard API, so any capabilities exposed by the API can be used. For anything not covered by the model card, test the model with your own prompts and payloads before putting it into production.
The best use cases come from the combination of multimodal input and large context. You can ask questions about a long video, summarize an audio recording, inspect images alongside supporting text, or work with file attachments in the same conversation. The 1,048,576-token context window also supports long documents and multi-file corpora. Use Muse Spark 1.2 when a task would otherwise require multiple models or many chunked calls. If the input is mostly text and short, a cheaper model may provide better value. OrcaRouter bills at the provider rate, so cost is proportionate to the tokens you send. To keep cost manageable, pre-filter content and only send the portions of files, images, or audio that are relevant. This model is not designed for generating media; its output is text.
Muse Spark 1.2 is positioned as a high-context multimodal model. For short text-only queries, a smaller model will typically be faster and less expensive. The provided pricing of $4.25 per 1M output tokens is a premium rate compared to many lightweight text models, though exact comparisons depend on the alternatives. There are no specific token budgets or benchmark numbers in the supplied facts, so the decision is based on task complexity. If your content is plain text under a few thousand tokens, use a cheaper model. If your task involves long audio, video, images, or files, the multimodal input support and 1M context can justify the price. OrcaRouter lets you switch models by changing the model parameter, so you can route simple jobs to a cheaper model and reserve Muse Spark 1.2 for harder multimodal tasks.
The model outputs text. It does not generate images, video, or audio. It accepts those media as inputs, but the response is a textual completion. There is no stated output token limit in the supplied metadata, but standard API settings such as max_tokens can cap the length of a response. Because no benchmark figures are provided, accuracy and capability claims should be verified with your own data. The model may not be able to process extremely long videos if the representation exceeds the 1,048,576-token context window. It is also not guaranteed to perform well on tasks outside the expected multimodal understanding space. When using OrcaRouter, the API returns errors for requests that exceed the context budget or violate provider input constraints.
The supplied model facts do not include any benchmark scores or evaluation tables for Muse Spark 1.2. This does not mean the model is untested; it means the metadata provided for this catalog entry does not list those numbers. To evaluate the model, run your own test examples and compare outputs on task-specific metrics. OrcaRouter does not publish benchmark results for third-party models in this entry. Without benchmark data, readers should rely on qualitative capabilities such as a 1,048,576-token context window and multimodal input support. Any performance numbers you see elsewhere should come from Meta directly or from independent evaluations. The phrase Muse Spark 1.2 should not be associated with specific benchmark claims here. If you need a quantitative quality measure, design a small evaluation set that covers your expected input types, including text, image, video, file, and audio. Measure task accuracy, output consistency, and cost. The absence of published benchmarks also underscores the need for testing before production deployment.
The context window is 1,048,576 tokens. This is the maximum input size the model can handle in one call, as stated in the provided facts. A token is a unit of text processing; 1,048,576 tokens is roughly a large amount of textual data, although exact words per token vary. This context limit also applies to content drawn from video, image, file, and audio inputs once they are converted into tokens. A large context allows the model to see an entire document, long transcript, or many source files without chunking. It also means you must be aware of token usage when uploading media files. If the total token count of a request exceeds the context window, the API call will fail. Use OrcaRouter's token counting tools or your own estimation to stay within limit.
No latency or throughput figures are included in the supplied metadata. The actual response time depends on several variables: total input length, number of modalities, complexity of the prompt, and provider-side load. A request involving video and audio may be slower than a text-only request because the associated data must be processed before inference. Output length also affects time; generating many tokens takes longer than generating a few tokens. There is no guaranteed speed or availability level stated for this model. To get reliable latency measurements, benchmark the API through OrcaRouter with representative requests. If you need extremely low latency for simple prompts, use a smaller model. For long-context multimodal processing, latency is likely higher by nature of the workload.
OrcaRouter bills Muse Spark 1.2 at $1.25 per 1,000,000 input tokens and $4.25 per 1,000,000 output tokens. These figures come from the provided facts and are passed through at the provider rate with zero markup. Input tokens are tokens sent by the client, including text, file content, and any image, video, or audio representation. Output tokens are the tokens returned in the generated text. The total cost of a call is therefore input tokens times $1.25/1M plus output tokens times $4.25/1M. Because OrcaRouter adds no markup, this is the line-item rate you should use for budgeting. The context window is 1,048,576 tokens, so a full-context input would cost slightly more than $1.25 at the input rate. That estimate assumes exactly the maximum input tokens; actual requests vary.
Input tokens cost $1.25 per million. Output tokens cost $4.25 per million. Output tokens are priced higher than input tokens, which is common for generative models. For many tasks, the output is a short completion while the input contains a large amount of source material. In that case input token cost dominates. Conversely, if you ask for a long generated text, output cost can become significant. Since the context window is 1,048,576 tokens, it is possible to send near that limit and incur input charges. You can reduce input cost by sending only relevant sections of files, limiting video or audio length, and trimming conversation history. OrcaRouter does not add a markup, so any difference in cost comes from usage volumes and choices.
The supplied facts do not include any cache pricing or discount information for Muse Spark 1.2. Some providers offer lower prices for cached input tokens, but no such figures are listed here. If OrcaRouter or Meta enables prompt caching for this model, the available cache pricing would be documented separately. Without that documentation, assume that every input token is billed at the full $1.25 per million rate. Do not plan your budget around cache savings unless you have confirmed the current terms. Caching also depends on prompt reuse; requests with unique or highly variable inputs will not benefit. Check OrcaRouter's pricing page or contact support to see if cache pricing exists for meta/muse-spark-1.2.
OrcaRouter states that Muse Spark 1.2 is billed at the provider rate with zero markup. That means the per-token price you pay matches the provider's listed rate: $1.25 per 1M input tokens and $4.25 per 1M output tokens. OrcaRouter is not adding a per-token fee or a service fee on top of this entry. A zero markup rate simplifies cost projection because your invoice line matches the model's list price. It does not mean there is no base subscription or infrastructure fee if OrcaRouter charges one for the API account; the provided facts only describe per-token billing. Always check your OrcaRouter plan for any fixed costs. The zero markup applies to the model rate and is the basis for cost comparisons with other models.
To call Muse Spark 1.2 through OrcaRouter, set the base URL to https://api.orcarouter.ai/v1 and use the model id meta/muse-spark-1.2. Your client should send a request that follows the OpenAI-compatible chat completion format. Include your OrcaRouter API key in the Authorization header. The messages array can contain the content you want to process. For multimodal input, the content format for image, video, file, and audio depends on OrcaRouter's supported schema; the OpenAI-compatible interface typically permits content parts. The exact field names are documented in the API reference. Make sure the total token count of the request stays within the 1,048,576-token context window. If you use the OpenAI SDK, set the base_url and model parameters accordingly.
OrcaRouter exposes an OpenAI-compatible API, so standard inference parameters like temperature, top_p, max_tokens, and stop sequences may be available. Because these are general OpenAI API parameters, their behavior follows the API specification. The provided facts do not list model-specific parameters for Muse Spark 1.2. You can also pass a system message, user messages, and assistant messages as part of the conversation. For multimodal input, the request format should include the correct content parts for the included media. There is no guarantee that every OpenAI parameter is supported by every provider, so test each parameter when you migrate. Use the model id meta/muse-spark-1.2 exactly. If a parameter is unsupported, OrcaRouter will return an error message; verify against the API documentation.
To migrate an existing OpenAI-compatible integration to OrcaRouter, change the base URL from your current endpoint to https://api.orcarouter.ai/v1 and set the model to meta/muse-spark-1.2. Keep the same message structure if you were using standard OpenAI chat completions. Replace your API key with an OrcaRouter key. If your existing requests include model-specific extensions, remove or adapt them based on OrcaRouter's API docs. The context window is 1,048,576 tokens, which may be larger than your previous model, allowing the same request to fit without truncation. Check your input formatting for image, video, file, and audio if you used provider-specific fields. Test with a small request first. Because Muse Spark 1.2 outputs text, migration is straightforward for text-based tools.
The supplied model facts do not contain data handling, retention, or privacy terms for Muse Spark 1.2 or OrcaRouter. In general, requests are sent to the model provider to generate completions. OrcaRouter may store API logs or usage data for operations, but the specific policy is not described here. If data privacy matters, review OrcaRouter's terms of service and privacy policy, and confirm whether Meta's provider terms apply to content you send. Do not assume that input video, audio, or files are automatically deleted after inference. For confidential data, consider redaction and local pre-processing. This section should not be read as legal advice. Contact OrcaRouter support to obtain the current data processing agreement. Your use of the API constitutes agreement to the applicable terms.
Muse Spark 1.2 differs from text-only language models because it accepts image, video, file, and audio inputs. A text-only model can only process written content. For multimodal tasks, Muse Spark 1.2 removes the need to extract and transcribe media before calling an LLM. Its 1,048,576-token context window also allows it to hold much longer inputs than many standard models. On the other hand, text-only models are often less expensive and may execute faster for plain-text work. The output price of $4.25 per million tokens is higher than many lightweight text models. If your project is entirely text and does not need long context, a text-only model might be a better fit. If your workflow already sends images, audio, or video, this model offers a single-call path.
Audio-only and image-only models typically specialize in one modality and may output labels, embeddings, or transcriptions. Muse Spark 1.2 unifies text, image, video, file, and audio in one text-generation interface. This can simplify pipelines that previously combined an audio transcriber, an image captioner, and an LLM. However, specialist models can be better tuned for tasks like speech recognition or object detection. No benchmark scores are provided for Muse Spark 1.2, so there is no data here to compare against specialist models. The appropriate choice depends on whether you need general understanding or dedicated accuracy. For general reasoning over multiple modalities, a single multimodal model can be easier to maintain. For strict production accuracy on one media type, test specialist alternatives.
The only pricing figures supplied for Muse Spark 1.2 are $1.25 per 1M input tokens and $4.25 per 1M output tokens, billed with zero markup through OrcaRouter. Comparisons to other models would require their pricing lists, which are not part of this entry. Generally, multimodal models are more expensive to run than simple text models because they process richer inputs. The 1,048,576-token context window can increase cost if you send large amounts of media. For a short text prompt, the input cost is very small. For a request with several minutes of video and a long output, the cost will be higher. Use the per-token rates and your expected token counts to estimate whether an alternative model would be more economical. Since OrcaRouter adds no markup, the provider rate is the relevant benchmark.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key=os.environ["ORCAROUTER_API_KEY"],
)
response = client.chat.completions.create(
model="meta/muse-spark-1.2",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)include_reasoningmax_tokensreasoningreasoning_effortrepetition_penaltyresponse_formatstructured_outputstemperaturetool_choicetoolstop_ktop_p| Input / 1M tokens | $1.25 |
| Output / 1M tokens | $4.25 |
| Cache read / 1M | $0.150 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
What developers are saying this week
GET /api/public/models/meta/muse-spark-1.2Open @misc{orcarouter_muse_spark_1_2,
title = {Muse Spark 1.2 API},
author = {Meta},
year = {2026},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/meta/muse-spark-1.2}
}Meta. (2026). Muse Spark 1.2 API. OrcaRouter. https://www.orcarouter.ai/models/meta/muse-spark-1.2