Amara S2 model card

The newest Amara model, available from the middle plan level up, with a larger output limit.

At a glance

Good for

• Long-form outputs that do not fit in 32,768 tokens: full reports, large code files, long translations
• Teams evaluating the next generation before it becomes the default

Consider another model for

• Production paths that need a published per-token price before launch; until the price row exists, credit-paid requests fall back to the plan’s flat pay-as-you-go rate.

Using it well

Run your evaluation set on S2 and S1.2 side by side before moving traffic; newer is not automatically better for every prompt.

Evaluation results

Syntic publishes benchmark results and safety evaluation summaries for each model on this page when they are complete. Until then, measure on your own task: the method in Choosing a model with 20 to 50 real inputs tells you more about fit than any public benchmark.

Safety and data handling

• Requests pass Syntic’s platform safeguards before reaching the model, and organizations can add their own inference policy hook.
• Prompts and outputs are not used to train models.
• Retention follows your organization’s data retention settings. Check Models overview for whether this model is available to zero-data-retention organizations.

Request

{ "model": "Amara S2", "messages": [{ "role": "user", "content": "Hello" }] }