Muse Spark 1.3

What is Muse Spark 1.3?

Muse Spark 1.3 is the newest version of Meta's closed Muse Spark model and the one Meta recommends for new work; versions 1.2 and 1.1 are still served. It reads text, images, video and PDFs, writes text, and holds just over 1M tokens of context. It always reasons, at six effort levels up to max. You reach it through the Meta Model API, which accepts OpenAI and Anthropic SDK calls, through Meta's Muse Code terminal agent, or on OpenRouter.

The twist is the price: the same model costs $1.25 / $4.25 per million tokens, or $0.10 / $0.20 if you let Meta train on your prompts and outputs.

What changed from Muse Spark 1.2?

Mostly agent behaviour and efficiency: it uses fewer tokens, adds a max reasoning level and scores far higher on long context, at the same price.

Efficiency

Meta engineers measured about 20% fewer tool calls and 25% fewer tokens on coding work, with shorter answers.

Max reasoning

A new max effort level goes beyond xhigh, on the Standard tier only.

Agent behaviour

It asks when a prompt is unclear, confirms before actions it cannot undo, and keeps several tasks apart in one thread.

Long context

It scores 98.1 on MRCR at 512K to 1M tokens, up from 55.5.

Price

Unchanged: both cost $1.25 / $4.25 per million tokens, or $0.10 / $0.20 on the Contributor tier.

Audio

Audio understanding is only partly supported in 1.3, so Meta suggests 1.2 for audio.

Who is Muse Spark 1.3 best for?

Muse Spark 1.3 is best for long agent tasks, coding in the terminal, cheap prototypes you are happy for Meta to train on, and work that needs a 1M-token context.

• Long agent tasks

  It is trained to track many jobs in one long thread, ask when a request is unclear and check with you before an action it cannot undo.

• Coding in the terminal

  Muse Code, Meta's multi-agent coding CLI, is built around it, and Meta says it uses about 20% fewer tool calls and 25% fewer tokens than version 1.2.

• Cheap prototypes

  The Contributor tier cuts the price to $0.10 per million input tokens if you are fine with Meta training on that traffic.

• Large files and codebases

  The 1,048,576-token context takes text, images, video and PDFs at one flat rate, and it scored 98.1 on Meta's MRCR test at 512K to 1M tokens.

What's included with Muse Spark 1.3?

Six reasoning levels, a cheaper Contributor tier, OpenAI and Anthropic compatible endpoints, web search, computer use and automatic prompt caching.

Six reasoning levels

Effort runs from minimal to max; max is only on the Standard tier, and reasoning cannot be switched off.

Contributor tier

The muse-spark-1.3-contributor model ID cuts token prices by more than 90% in exchange for training rights, at 100 requests a minute instead of 3,000.

OpenAI and Anthropic formats

The API speaks the Responses, Chat Completions and Messages formats, so Claude Code, Codex and OpenCode connect with a base URL and a key.

Web search grounding

A built-in web_search tool adds live results for $2.50 per 1,000 searches on top of token costs.

Computer use

A native computer tool on the Responses API returns mouse and keyboard actions for a browser or desktop you run yourself.

Automatic prompt caching

Repeated prompt prefixes are cached with no setup and billed at $0.15 per million tokens, or $0.002 on the Contributor tier.

How much does Muse Spark 1.3 cost?

Muse Spark 1.3 costs $1.25 per million input tokens and $4.25 per million output tokens on Meta Model API, or $0.10 and $0.20 if you let Meta train on your prompts, and Muse Code subscriptions start at $5 a month.

Power Usage is $50 a month for 20 times the Everyday Usage limit, early access to new features and bigger file uploads. Muse Code plans work only inside the Muse Code terminal agent, and Meta says their availability varies by region; any other API key is billed per token. Web search grounding adds $2.50 per 1,000 searches, and there is no extra charge for long prompts across the 1M-token context. The API has no minimum spend. Full pricing on dev.meta.ai. We checked these prices on 28 Sep 2026.

How do readers rate Muse Spark 1.3?

No one has rated Muse Spark 1.3 yet. Be the first.

Muse Spark 1.3 pros and cons

It scores at or near the top of Meta's own coding and long-context table at a low price, but the cheapest tier means Meta trains on your data and real-world reports are still thin.

What are the best Muse Spark 1.3 alternatives?

Four tools on Recaply do Muse Spark 1.3's job. DeepSeek V4.1 Flash, GLM-5.3-Flash and Ornith-1.0 are open source, so you can run them on your own server.

Gemini 3.8 FlashFree

Choose it when you want a free tier to prototype on or half-price batch jobs at $0.75 / $3.75 per million tokens; its normal paid rate of $1.50 / $7.50 is above Muse Spark's Standard tier.

DeepSeek V4.1 FlashOpen Source

Choose it when you want MIT-licensed open weights with a 1M context, or an API at $0.30 / $1.20 per million tokens at peak and half that off-peak.

GLM-5.3-FlashOpen Source

Choose it when you want a flat coding plan that works in Claude Code, Cline and OpenCode, from $18 a month; Muse Code plans only cover Meta's own CLI.

Ornith-1.0Open Source

Choose it when long-running coding agents need open weights in a size that fits your hardware.

Questions about Muse Spark 1.3