Muse Spark 1.3 API: Meta's flagship is now on FastMetal
Meta's Muse Spark 1.3 is available on FastMetal as muse-spark-1.3. It follows Muse Spark 1.2 (muse-spark-1.2), which stays in the catalog.
What you get
- A 1,048,576-token context, the same as 1.2. A long spec or a whole codebase fits in one request.
- The same per-token rates as 1.2. Meta charges the same for both versions, so we bill them the same, and moving up a version does not raise the bill.
- One upstream endpoint: Meta's own. Requests are not spread across hosts that run the model at different quantizations, so output quality stays consistent from run to run.
Leave headroom in max_tokens
Muse Spark 1.3 reasons before it answers, and those reasoning tokens count against max_tokens.
We asked it to reply with just "ok" and set max_tokens to 200. It spent 190 of those tokens on reasoning. With a smaller limit, it runs out mid-reasoning and returns an empty message.
Give it a few hundred tokens at least, even for short answers. If you get empty replies, raise max_tokens before you suspect the model.
Calling it
Set model to muse-spark-1.3. The base URL and API key are the same as for every other model on FastMetal:
curl https://api.fastmetal.ai/v1/chat/completions \
-H "Authorization: Bearer $FASTMETAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-spark-1.3",
"messages": [{"role": "user", "content": "List three contradictions in this spec."}],
"max_tokens": 2000
}'
With the OpenAI SDK, point base_url at https://api.fastmetal.ai/v1 and change model. Nothing else changes. To try it without code, pick it in chat.
1.2 or 1.3?
The rates are the same, so cost does not decide it. Compare the output instead: Muse Spark 1.2 vs Muse Spark 1.3 shows both models' real answers to the same prompts. For how each one ranks today, see the rankings by use case.
Pricing
Billed per million input and output tokens, from a prepaid balance, with no monthly fee. Current rates are on pricing and the model pages in the catalog.