Muse Spark 1.2 Limits

Muse Spark 1.2 limits were not published in full at Meta's 2026-08-05 launch. This page states what Meta confirmed, what is still unknown, and where to verify.

The rule for this page is strict: nothing is invented. If Meta has not published a number, you will see the phrase "not published — verify at https://developer.meta.com/ai/products/muse-code/" and a suggestion for what to check.

Use this as a pre-integration checklist before wiring Muse Spark 1.2 into a production system.

Rate Limits (RPM / TPM)

Meta has not published requests-per-minute or tokens-per-minute limits for Muse Spark 1.2 at launch. Verify at https://developer.meta.com/ai/products/muse-code/.

For any production workload — batch scoring, high-QPS agent loops, or CI-integrated code review — you should confirm both per-key and per-org limits directly before capacity planning.

Muse Code inherits these limits since it uses Muse Spark 1.2 as its backing model. Multi-agent runs can burn through rate quotas faster than single-turn chat because workers and reviewers issue requests in parallel.

• RPM: not published — verify at vendor page
• TPM: not published — verify at vendor page
• Multi-agent workloads amplify rate consumption
• Confirm per-key AND per-org limits

Context Window and Output Cap

Muse Spark 1.2 context window and output-token cap were not published at launch — verify at https://developer.meta.com/ai/products/muse-code/.

Muse Spark 1.1 shipped with a 1M-token context window. Whether 1.2 preserves that, expands it, or trades depth for cost is not documented in the launch materials.

For long-horizon coding sessions (large repos, long test outputs, replay of full event logs), a 200k+ window is the practical floor. Confirm before you architect a repo-wide agent.

• Context window: not published — verify at vendor page
• Output-token cap: not published
• 1.1 baseline was 1M tokens (unconfirmed for 1.2)
• Repo-wide agents need 200k+ minimum

Modality and Built-In Tools

Muse Spark 1.2 is positioned as coding-grade, but Meta has not documented vision, audio, or structured-data modality beyond text and code — verify at https://developer.meta.com/ai/products/muse-code/.

Muse Code's exact list of built-in tools was not published beyond the "file edit, tool call" bucket. Whether it includes shell exec, web fetch, git ops, or MCP support is not documented in the launch materials.

Fine-tuning availability for Muse Spark 1.2 is also unpublished. Teams that rely on domain fine-tunes should not assume parity with prior Meta open-weights releases.

• Vision / audio / structured-data: not published
• Full built-in tools list: not published
• Fine-tuning: not published
• Do not assume open-weights parity

Compliance, SLA, and Data Residency

Meta has not published enterprise SLA, SOC 2, HIPAA, or data-residency status for Muse Spark 1.2 at launch. Verify at https://developer.meta.com/ai/products/muse-code/.

The Contributor tier explicitly permits Meta to use prompts and outputs to improve their products. Regulated teams should default to Standard until Meta publishes formal certifications.

In our engagement with HOA and condo boards we treat every new model as SOC-2-absent until the vendor page shows an actual attestation. Muse Spark 1.2 is at that stage today.

• SLA: not published
• SOC 2 / HIPAA: not published
• Data residency: not published
• Regulated teams default to Standard, not Contributor

How to use Muse Spark 1.2

You do not host Muse Spark 1.2 yourself — you use it through a tool, so "getting started" really means choosing the right one.

The fastest way to put Muse Spark 1.2 to work day to day is inside an AI IDE, and Cursor is the most popular — it supports it directly, so you can be working in minutes. Prefer a different editor? Windsurf, Zed, and GitHub Copilot drive these models too.