Muse Spark 1.3 launches with a Standard and a data-for-discount Contributor tier
Muse Spark 1.3 is Meta’s first-party Model API release for long-horizon coding and agentic work, sold in two tiers on the same model: Standard at $1.25/M input and $4.25/M output, and a Contributor tier priced up to 95% lower in exchange for Meta training on your inputs and outputs. Both share a 1M-token context window and native tool calling and MCP support.
By Todd · Published 2026-08-14 · Verified 2026-09-02
| Endpoint | muse-spark-1.3 / muse-spark-1.3-contributor |
|---|---|
| Context | 1,048,576 input tokens |
| Max output | 128,000 tokens |
| Standard price | $1.25/M input, $0.15/M cached input, $4.25/M output |
| Contributor price | $0.10/M input, $0.002/M cached input, $0.20/M output |
| Verified | 2026-08-14 |
What Muse Spark 1.3 is
Muse Spark 1.3 is Meta’s proprietary model served through a first-party, OpenAI-compatible Model API — a break from the open-weight Llama models Meta previously shipped for third-party hosts. Meta positions it for long-horizon coding and multi-step agentic work, with improved tool, browser, and computer use across harnesses, native tool calling, and MCP support.
Two tiers, one model
Meta sells the same model and capability set at two price points. The Standard tier keeps your inputs and outputs private, priced at $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens. The Contributor tier is up to 95% cheaper — $0.10/M input, $0.002/M cached input, and $0.20/M output — because Meta uses your inputs and outputs to train and improve its models, and the tier is additionally rate-limited by tokens.
- Standard: private, full-price, no published rate limit beyond account defaults
- Contributor: same model, steep discount, trains on your traffic, token rate-limited
- Selecting a tier is a model-ID choice, not a request flag
Who should use which tier
Contributor makes sense for cost-sensitive, non-sensitive workloads — internal tooling, public-data agents, high-volume experimentation — where the training trade-off is acceptable. Standard is the only option once a workload touches proprietary code, customer data, or anything else you would not want used as training data.
Migration and integration notes
The API is OpenAI-compatible at api.meta.ai/v1, so existing OpenAI SDK integrations mostly need a base-URL and model-ID change. There is no published reasoning-effort parameter, structured JSON-schema mode, or seed control — only basic json_object output — so teams migrating from providers that rely on those knobs should re-test their JSON-parsing and determinism assumptions.
Known limits and open questions
Muse Spark 1.3 has no graded score yet on All AI Ask’s coding test, so this page does not crown it an overall winner — only a price-and-context comparison. Check Meta’s current pricing page before budgeting, since Contributor-tier rate limits are token-based and can change.
Explore the canonical data
Test it yourself
Run the same prompt across muse spark 1.3, muse spark 1.3 contributor, claude sonnet 5 in one workspace.
Open the comparison playground →FAQ
What is the difference between Muse Spark 1.3 and Muse Spark 1.3 Contributor?
They are the same model and capabilities. Contributor costs up to 95% less because Meta trains on your inputs and outputs, and it is additionally rate-limited by tokens; Standard keeps your data private at full price.
What context window does Muse Spark 1.3 support?
Both tiers share a 1,048,576-token (1M) input context window and a 128,000-token maximum output, per Meta’s pricing documentation.
Is Muse Spark 1.3 OpenAI-SDK compatible?
Yes. Meta serves it through an OpenAI-compatible chat/completions endpoint at api.meta.ai/v1.