Amara Deep model card

For problems that take several steps of reasoning to get right: planning, multi-constraint decisions, careful analysis.

At a glance

Good for

• Analysis where a wrong answer is costly: contracts, financial reconciliation, root-cause investigations
• Planning steps for agents
• Grading other models’ outputs in evaluations

Consider another model for

• Latency-sensitive chat, where its extra thinking time shows.
• Simple tasks where Amara Quick already passes your evaluation.

Using it well

Give it the whole problem and the success criteria up front rather than steering it turn by turn; planning ahead is where it shines.

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 Deep", "messages": [{ "role": "user", "content": "Hello" }] }