Multi-model responses, finer injection controls, flexible pricing tiers

Multi-model panels with a judge model now supported in Merge Fusion

New

You can now send one prompt to a panel of models in parallel, then have a judge model synthesize their answers into a single response.

• Choose both the panel and the judge yourself
• An all open-source panel matched Fable 5 on the DRACO benchmark at a quarter of the cost
• A premium panel beat Fable 5 outright by 8.5 points

Read more in the Merge Fusion docs.

Per-axis modes and thresholds now supported for indirect prompt injection protection

Security

You can now set indirect prompt injection protection per axis, picking a mode (off, alert, or block) and a sensitivity threshold for each one.

Test the configuration against injection patterns with a built-in detection tester in Security settings before it goes live. Read more in the prompt injection protection docs.

Flex and priority service tiers are now selectable per request

New

You can now set a service_tier request parameter to toggle between flex and priority processing, trading cost for latency or the reverse.

Per-tier pricing and billing now show up directly on /v1/models, so the tradeoff is visible before you pick a tier. Read more in the service tiers docs.