Meta Model API vs Beam 1.0

Meta Model API

Meta Model API serves Muse Wonder, in preview since 12 October 2026, and Muse Spark with 1M-token context, constrained JSON Schema output, automatic prompt caching and per-token prices published without a login. Standard-mode content is not used for training. There's no free allowance, status history, OpenAPI file, deprecation policy or way to force a tool call.

Beam 1.0

The Reflection API is OpenAI-compatible, with an OpenAPI document, strict structured outputs and tool calls, and error codes that say when to retry, and the weights are Apache-2.0. It is a free public beta with unpublished daily limits and no per-token price, reasoning can't be turned off, and three API incidents were posted between 14 and 16 October 2026.

Meta Model API

1. Use muse-wonder-1.0 for long multi-step work and muse-spark-1.3 for cheaper, high-volume calls. Both take the same requests
2. Keep personal or confidential data away from -contributor model IDs. Meta may train on that content
3. Honour Retry-After on 429 and 503, which is 60 seconds for service_overloaded, and check GET https://api.meta.ai/v1/status, which needs no key
4. Pass the key to the OpenAI SDK explicitly or set OPENAI_API_KEY. The SDK doesn't read MODEL_API_KEY
5. Leave tool_choice at auto. The docs say none and required are not supported

Beam 1.0

1. Send reasoning_effort low or medium for short tasks. The default is max, and none or minimal return 400 unsupported_value
2. Return every assistant message with its reasoning_content unchanged. If a turn comes back without it, start a new conversation, since later requests fail with missing_required_parameter
3. Don't rely on stop or seed. Both are accepted and ignored, so truncate output in your own code
4. Honour x-should-retry: false on a 429. It means a daily limit is spent and Retry-After counts to 00:00 UTC
5. Set max_completion_tokens yourself on long prompts. The default budget isn't reduced to fit the prompt and includes reasoning tokens