Move to Amara S2
Amara S2 is the newest Amara model. This guide lists the differences from Amara S1.2 and how to make the move safely.
What is different
| Amara S1.2 | Amara S2 | |
|---|---|---|
| Maximum output | 32,768 tokens | 65,536 tokens |
| Context window | 262,144 tokens | 262,144 tokens |
| Plan level | Every level with API access | Middle (Starter, Business, Pro) and above |
| Price | 2.50 / 0.40 / 12.00 | Not yet published; credit-paid requests use the plan’s flat pay-as-you-go rate meanwhile |
| Default model | Yes | No |
Steps
Check access
GET /v1/models with your key should list Amara S2. If it does not, your plan level is below the minimum.
Run your evaluation on both
Send the same 50 to 500 cases through both models with the same prompt, using an evaluation run with two variants. Compare pass rate, output length and cost.
Revisit max_tokens
If you had to split long outputs because of the 32,768 limit, you can now ask for up to 65,536. Set max_tokens deliberately; a higher ceiling also allows longer, costlier replies.
Adjust prompts only where results say so
Newer models often follow instructions more literally. If replies became terser or more rigid, loosen over-specified rules rather than adding new ones.
Shift traffic gradually
Route 5%, then 25%, then 100% of requests over several days, watching error rate, cost per request and user feedback at each step.
Rolling back
Keep the model name in configuration. Rolling back is a configuration change, not a deploy.