The developer of the Ampersand assistant updated its Agent API this week to include a new high-capacity tier and lower fees for existing service levels. The company is reducing costs for all of its “effort levels,” which are bundled agent configurations optimized for specific research tasks, and has introduced an “ultra” level to handle the most demanding and long-running requests.
The pricing adjustments are part of an effort to optimize the API’s technology stack to maximize the amount of intelligence provided per dollar. The updates follow the release of Ampersand’s multi-agent coding mode, which recently recorded a 92.2% score on the SWE-bench Pro v2 HARD-51 benchmark, matching the performance of Fable 5.1 and Gemini 4 Argon.
