Muse Spark 1.3

Access Meta’s Muse Spark 1.3 API on GPTProto for coding assistants, text analysis, and multi-step agent workflows. The model combines a documented 1M-token context window with tool calling and adjustable reasoning. Use one API key and one balance across 200+ models to evaluate the right fit for your application.

モダリティ

API 利用例

curl --request POST "https://gptproto.com/v1/chat/completions" \
    --header "Authorization: Bearer $GPTPROTO_API_KEY" \
    --header "Content-Type: application/json" \
    --data '{
        "model": "muse-spark-1.3",
        "messages": [
            {
                "role": "user",
                "content": "Hello"
            }
        ]
    }'

Muse Spark 1.3 料金

チャット、コーディングエージェント、ドキュメント作業。1Mトークン単位の料金 — 入力、キャッシュ入力、出力は別々に課金。GPTProtoは公式料金より20%低価格です。

シナリオ	Meta リスト	OpenRouter	GPTProto	月間の節約
個人10M トークン / 月（キャッシュ 4.8M）	$13.22	$13.95	$10.58	−$2.64≈ $31.73 / 年
チーム100M トークン / 月（キャッシュ 48M）	$132.20	$139.47	$105.76	−$26.44≈ $317.28 / 年
ビジネス500M トークン / 月（キャッシュ 240M）	$661.00	$697.35	$528.80	−$132.20≈ $1586.40 / 年

提供元別の月額コスト

OpenRouter

$0.00

Muse Spark 1.3 API

Build text-to-text applications that turn requirements, source code, and long conversations into useful answers and next actions. Muse Spark 1.3 is designed for work that involves multiple steps: investigating a bug, comparing evidence, revising a plan, and continuing after a tool returns new information.

1M-Token Model Context

Meta documents a 1,048,576-token context window for combining source text, instructions, and conversation history. Input and generated output share that budget, so leave room for the answer.

Coding Across Multiple Steps

Use the model in an agent that inspects code, proposes changes, runs tests, and revises a fix. Your application supplies the file access and execution tools.

Adjustable Reasoning Effort

Meta’s Standard model supports six reasoning levels, from minimal to max. Match effort to task difficulty; additional reasoning consumes output tokens and can increase response time.

Tool Calls and Structured Results

The model supports function calling and schema-based JSON output. Build workflows that retrieve information, classify requests, or return fields your application can process, using the options supported by your API route.

What Is Muse Spark 1.3?

Muse Spark 1.3 is Meta’s reasoning model for coding and agentic work, released on September 2, 2026. Its training emphasizes longer tasks, following detailed instructions, and using intermediate results to decide what to do next. This page covers text input and text output for applications built through GPTProto.

Model detail	What it means for your application
Developer	Meta
Official Standard model ID	`muse-spark-1.3`
Main-page task	Text to text
Model context window	1,048,576 tokens, shared by input and output
Response types	Text, code, and structured text such as JSON
Agent capabilities	Tool calling and iterative reasoning within a connected application

These are model-level specifications. Use GPTProto’s API Usage example for the exact request model ID, endpoint, and options available through your account.

Muse Spark 1.3 API Applications

Coding assistants and maintenance. Give an assistant a bug report, relevant source files, and failing test output. Ask it to trace the cause, propose a patch, and explain the change. With execution tools connected, it can use test results to revise its work before presenting a final diff.

Research and business workflows. Turn supplied customer feedback, product requirements, or campaign notes into an evidence-backed brief. Ask for a recommendation, supporting excerpts, and unresolved questions. For workflows that also retrieve records or update systems, connect the relevant tools through your agent application.

Text processing inside apps. Extract fields from support requests, classify incoming messages, or turn a long conversation into a handoff record. Define the output fields before running the task, then validate the returned data before passing it to another system.

In Meta’s internal engineering comparisons, version 1.3 used approximately 20% fewer tool calls and 25% fewer tokens than 1.2. Those reported results suggest a useful evaluation target: measure the total cost of a completed task, including retries. They are not a guaranteed saving for every workload.

Muse Spark 1.3 Contributor vs Standard

The Contributor variant changes the conditions under which you use the model. Meta offers lower token prices in exchange for permission to use submitted content and responses for training. It also limits the available reasoning levels.

Difference on Meta Model API	Standard	Contributor
Official model ID	`muse-spark-1.3`	`muse-spark-1.3-contributor`
Content used to train Meta models	No, under Meta’s Standard-mode policy	May be used for training
Maximum reasoning effort	`max`	`xhigh`; `max` is unavailable
Pricing arrangement	Standard rates	Discounted rates with the data-use trade-off

Source: Meta’s pricing modes and reasoning documentation. This describes Meta’s variants; it does not establish a separate Contributor offering or a data-retention promise on GPTProto. Check the selected route’s terms when evaluating sensitive workloads.

Plan Context and Reasoning for Long Tasks

A large context window helps keep relevant material together, but an agent still needs conversation management. Preserve the task’s requirements and recent results; summarize older steps when their exact wording is unnecessary. For code review, keep the original code available where precise details matter.

Reasoning tokens count toward generated output, even when they are absent from the visible answer. Start with a moderate effort level for routine work and compare higher effort on difficult cases. When connecting an existing client, check supported reasoning, tool, and response-format options alongside the base URL and credentials.

Muse Spark 1.3 API FAQ

How much does Muse Spark 1.3 cost on GPTProto?

Use the current input, output, and applicable cache rates in this page’s pricing panel. For agents, compare the cost of completing the whole task: repeated prompts, reasoning, retries, and final answers all affect usage. See GPTProto pricing for the platform’s current rates.

Which API key should I use to access the model?

Requests through GPTProto use a GPTProto API key. Copy the endpoint and model value from this page’s API Usage example. Meta’s direct API credentials and provider-specific request options are separate; check those details when moving an existing integration.

Can I test prompts in a playground before integration?

Use the page’s Playground entry to evaluate representative text tasks, such as reviewing a short function or summarizing a support conversation. Test tool execution separately in your application: a chat response alone does not demonstrate that an entire agent workflow works.

Is the context window also the maximum answer length?

No. The documented 1,048,576-token window is a combined budget for input and generated output. Conversation history and tool results consume space too. Reserve output capacity before sending a large request, and use the output limit supported by your GPTProto route.

Is Muse Spark 1.3 open source or available for self-hosting?

Meta currently documents Muse Spark as a hosted model. Its separately downloadable Muse Glimmer model is distilled from Spark, but it is a different model. API access to Spark 1.3 does not provide downloadable Spark weights or a self-hosting license.

How should I compare it with Gemini 4 Argon, GPT-6 Astra, or Claude Opus 5.5?

Run the same tasks with equivalent instructions, tools, and acceptance criteria. Compare correct results, completion time, retries, and total spend. For coding, use passing tests and reviewable changes; for analysis, check evidence and instruction following. Explore the model catalog for available alternatives.

Can I use the model with images, video, files, or audio?

For other input types, see the dedicated image-to-text, video-to-text, and file-to-text pages. For audio-to-text support and limitations, note that Meta says audio understanding in version 1.3 is not fully supported and results may be degraded. A transcription-first workflow can pass transcript text into this main-page workflow.

このモデルに関連するガイド、比較、最新情報。

すべての記事

I Used Muse Wonder 1.0 API for Research Work: From Sources to an Action Plan

Use Muse Wonder 1.0 API to turn documents and research sources into verified decisions and tasks, with prompts, costs, safeguards, and a GPTProto trial.

Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5: Which AI Model Is Best?

Compare Gemini 4 Argon, GPT-6 Astra and Claude Opus 5.5 across official API pricing, coding benchmarks, context windows, output limits and computer-use capabilities. Explore model-specific API access through GPTProto with one key and a shared balance.

Top Chinese LLM Providers in 2026: 9 Companies Ranked for Developers

Kimi leads the Chinese Arena—but Qwen wins overall. Compare 9 Chinese LLM providers by API access, pricing, open weights, licensing, and production fit.

6 Best LLM API Providers in 2026: Multi-Model Platforms Compared

Compare the six best LLM API providers in 2026 by model access, pricing structure, speed, compatibility, governance, and developer experience.