Muse Spark Open vs SambaCloud

Muse Spark Open

Muse Spark Open runs on the owner's hardware in seven runtimes, with OpenAI-compatible servers in vLLM and SGLang, standard tool calls, a 1M-token context and Meta's own serving docs. Serving the 537 GB NVFP4 checkpoint takes four B300 GPUs, Meta's download is gated by manual review, and the licence adds attribution and a revenue threshold for model resellers.

SambaCloud

Per-token prices and the model list are readable without a key at /v1/models, and a free tier needs no card. Production models get two to three weeks' notice before removal, 13 model ids left between March and June 2026, and the docs disagree with the live catalogue on MiniMax-M2.7.

Muse Spark Open

1. Start vLLM or SGLang with --reasoning-parser muse --tool-call-parser muse (vLLM also --enable-auto-tool-choice), or tool calls come back as raw text
2. Set reasoning_effort to low or medium for short tasks. The default is xhigh, reasoning can't be turned off, and other values return an error
3. Stop only on <|eot|> and end of text. Stopping on <|eom|> cuts off tool calls and final answers
4. Stream long generations and leave max_tokens room for reasoning, or content can come back empty with finish_reason length
5. Don't call Meta Model API for this model. Use a self-hosted server or a hosted provider's model string, such as meta-models/muse-spark-open

SambaCloud

1. Call GET https://api.sambanova.ai/v1/models at start-up and use only ids it returns. The docs name models the endpoint no longer lists
2. Read x-ratelimit-remaining-requests and x-ratelimit-remaining-requests-day on every response. The free tier allows 20 requests a day per model
3. Treat 429 queue_full and 503 maintenance as retryable after a delay, and 410 model_deprecated as a signal to change model
4. Validate JSON output yourself. Schema enforcement is best effort and strict: true changes nothing
5. Check max_completion_tokens per model. DeepSeek-V3.1 caps output at 7,168 tokens and Llama 3.3 70B at 3,072