Model Selection & Options

A MAG model name has this form:

<base>[:<modifier>]...[@<provider>]

Example: gpt-6-luna:high@openai.

To find the base names, see Models tab.

Base name

The base name is a model name or an alias. An alias, such as claude-sonnet, points to the latest version in a model family. An exact version, such as claude-sonnet-5-5, does not change.

Modifiers

A modifier selects a variant of the model. Add each modifier after the base name, with a colon. Do not use spaces.

Dimension	Modifiers	Effect
Reasoning effort	none, minimal, low, medium, high, xhigh, max	Sets how much the model reasons before it answers
Service tier	flex	Uses the discounted tier. The request can take more time
Compression	compact, compressed, nocompress	Sets how much MAG shrinks the prompt

Use one modifier from each dimension at most. A model accepts only some of the modifiers. Example: gpt-6-luna:high:flex.

If a modifier is not valid for the model, MAG returns a 400 error with the code amsc_invalid_model_modifier.

Provider

The @<provider> suffix selects the company that serves the model. The suffix also selects the allocation bucket that MAG charges. Put the suffix last.

Provider	Source
openai	OpenAI
amazon	Amazon Bedrock
google	Google Vertex AI
microsoft	Microsoft Azure

Without a provider, MAG selects a provider that serves the model. MAG then keeps your requests on that provider.

Headers

Some applications let you set the model name only. For these applications, set a request header instead of a modifier.

Header	Values
X-AmSC-MAG-Provider	openai, amazon, google, microsoft
X-AmSC-MAG-Effort	none, minimal, low, medium, high, xhigh, max
X-AmSC-MAG-Service-Tier	flex, standard
X-AmSC-MAG-Compression	compact, compressed, off

A modifier in the model name overrides the header. MAG ignores a header value that the model cannot use.

Example:

curl https://mag.genesis.americansciencecloud.org/v1/chat/completions \
  --header "Authorization: Bearer $AMSC_TOKEN" \
  --header "X-AmSC-MAG-Provider: openai" \
  --header "Content-Type: application/json" \
  --data '{"model": "gpt-6-luna", "messages": [{"role": "user", "content": "Hello"}]}'

The same request on Microsoft Azure:

curl https://mag.genesis.americansciencecloud.org/v1/chat/completions \
  --header "Authorization: Bearer $AMSC_TOKEN" \
  --header "X-AmSC-MAG-Provider: microsoft" \
  --header "Content-Type: application/json" \
  --data '{"model": "gpt-6-luna", "messages": [{"role": "user", "content": "Hello"}]}'

Provider support notes

• Codex works with both openai and microsoft.
• Claude models are available on amazon and google. The web search tool is not supported on amazon and is supported on google.

Check what MAG used

Each response has the header X-AmSC-MAG-Applied. This header shows the provider, effort, tier, and compression that MAG used for the request.