Many people treat large models like chat boxes, but some tasks are better handled by asking a single structured question: yes or no, multiple-choice classification, scoring, or drawing a box around an object in an image. TypeSafe’s Jev (System One decision model) is built for exactly this—not as a chat substitute.
apikey.fan now has an official direct-connection group for TypeSafe, with both jev-1.14.0 and jev-1.13.0 available. The current price on the pricing page is about $0.033 per million input tokens, compared with the official $0.042—around 75–80% of the official price. Outputs are not billed. Version 1.14 adds the Box question type, which can return object coordinates. Both versions support images, and the gateway’s long context can be extended by default to around 260,000 tokens without increasing the per-token price.
What’s a more mature way to use it? The easiest option is Magpie: add TypeSafe to a routing group and select typesafe/jev-latest as the classifier. Have Jev determine intent first on each turn, send simple queries to a cheaper, faster model, and use Claude or Kimi for writing tests or tackling harder problems. The Routing page also shows confidence scores. The official recommendation follows the same approach: use choice or score as a gate, and only invoke the more expensive model when confidence is high enough. For more advanced workflows, you can send a batch of noul in a single request for jailbreak and harm screening, or use it to route support tickets and automatically categorize forms. When reviewing images with version 1.14, first ask whether there’s an object, then trust the Box coordinates.
By the way, the Kimi group on the same site already lists kimi-k3.1-pro, with discounted pricing of about $0.97 / $5.1 per million—well below the official $2 / $10.
To register, visit apikey.fan. See the site for documentation and instructions on using Magpie. The actual charge is based on the group’s rate multiplied by usage; prices on the page are rounded, so refer to your bill for the exact amount. This is a tools-focused share, not investment advice.
apikey.fan now has an official direct-connection group for TypeSafe, with both jev-1.14.0 and jev-1.13.0 available. The current price on the pricing page is about $0.033 per million input tokens, compared with the official $0.042—around 75–80% of the official price. Outputs are not billed. Version 1.14 adds the Box question type, which can return object coordinates. Both versions support images, and the gateway’s long context can be extended by default to around 260,000 tokens without increasing the per-token price.
What’s a more mature way to use it? The easiest option is Magpie: add TypeSafe to a routing group and select typesafe/jev-latest as the classifier. Have Jev determine intent first on each turn, send simple queries to a cheaper, faster model, and use Claude or Kimi for writing tests or tackling harder problems. The Routing page also shows confidence scores. The official recommendation follows the same approach: use choice or score as a gate, and only invoke the more expensive model when confidence is high enough. For more advanced workflows, you can send a batch of noul in a single request for jailbreak and harm screening, or use it to route support tickets and automatically categorize forms. When reviewing images with version 1.14, first ask whether there’s an object, then trust the Box coordinates.
By the way, the Kimi group on the same site already lists kimi-k3.1-pro, with discounted pricing of about $0.97 / $5.1 per million—well below the official $2 / $10.
To register, visit apikey.fan. See the site for documentation and instructions on using Magpie. The actual charge is based on the group’s rate multiplied by usage; prices on the page are rounded, so refer to your bill for the exact amount. This is a tools-focused share, not investment advice.
