sscoble/pi-mono
1
0
Fork 0
mirror of https://github.com/badlogic/pi-mono.git synced 2026-11-05 23:37:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

fix(ai): add Mantle strict mode and Daybreak models missing from models.dev

Browse source
This commit is contained in:
Cristina Poncela Cubeiro 2026-10-28 15:49:28 +01:00
parent bcbad6f8b1
commit bbde7ed20b
2 changed files with 50 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

46
packages/ai/catalog/providers/amazon-bedrock.ts
View file

@ -1,5 +1,5 @@
import type { Model } from "../../src/types.ts";
import { type ChatModel, defineProvider, modelsDevModels, withInputLimits } from "../define.ts";
import { addMissing, type ChatModel, defineProvider, modelsDevModels, withInputLimits } from "../define.ts";
import type { ModelsDevModel } from "../sources/models-dev.ts";
// Opus 5 is only callable through cross-region inference profiles.
@ -24,15 +24,47 @@ function mantleModel(model: ChatModel, source: ModelsDevModel): Model<"openai-re
api: "openai-responses",
// models.dev writes the region as a `${AWS_REGION}` template.
baseUrl: responses.baseUrl.replace(/\$\{AWS_REGION\}/, "{region}"),
// Mantle does not document OpenAI's extended prompt cache retention.
compat: { supportsLongCacheRetention: false },
compat: {
// Mantle does not document OpenAI's extended prompt cache retention.
supportsLongCacheRetention: false,
...(source.structuredOutput ? { supportsStrictMode: true } : {}),
},
};
}
/**
* Mantle-only models on AWS's model cards that models.dev does not list yet.
* Prices and limits come from the model cards:
* https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html
*/
const MANTLE_MODELS_MISSING_UPSTREAM: Model<"openai-responses">[] = [
{
id: "openai.gpt-5.6-cyber",
name: "Daybreak Red: GPT-5.6 Cyber",
cost: { input: 13.75, output: 82.5, cacheRead: 1.375, cacheWrite: 17.1875 },
contextWindow: 272_000,
},
{
id: "openai.gpt-daybreak-blue-5.6-sol",
name: "Daybreak Blue: GPT-5.6 Sol",
cost: { input: 5.5, output: 33, cacheRead: 0.55, cacheWrite: 6.875 },
contextWindow: 1_000_000,
},
].map((model) => ({
...model,
api: "openai-responses",
provider: "amazon-bedrock",
baseUrl: "https://bedrock-mantle.{region}.api.aws/openai/v1",
reasoning: true,
input: ["text", "image"],
maxTokens: 128_000,
compat: { supportsLongCacheRetention: false, supportsStrictMode: true },
}));
export default defineProvider({
id: "amazon-bedrock",
models: (ctx) =>
modelsDevModels(ctx, "amazon-bedrock", {
models: (ctx) => {
const models = modelsDevModels(ctx, "amazon-bedrock", {
api: "bedrock-converse-stream",
baseUrl: "",
include: (source) =>
@ -50,6 +82,8 @@ export default defineProvider({
...(source.structuredOutput ? { compat: { supportsStrictMode: true } } : {}),
};
},
}),
});
return addMissing(models, MANTLE_MODELS_MISSING_UPSTREAM);
},
finalize: (model) => withInputLimits(model, { images: { maxPerMessage: 20 } }),
});

10
packages/ai/test/bedrock-mantle.test.ts
View file

@ -32,6 +32,16 @@ describe("Bedrock Mantle", () => {
expect(converse.some((model) => model.id.includes("anthropic.claude"))).toBe(true);
});
it("lists Mantle models from AWS's model cards that models.dev does not", () => {
const ids = mantleModels().map((model) => model.id);
expect(ids).toEqual(expect.arrayContaining(["openai.gpt-5.6-cyber", "openai.gpt-daybreak-blue-5.6-sol"]));
});
it("uses strict tool schemas on Mantle models that support structured output", () => {
const model = mantleModels().find((candidate) => candidate.id === "openai.gpt-5.6-sol");
expect(model?.compat).toMatchObject({ supportsStrictMode: true, supportsLongCacheRetention: false });
});
it("sends Responses requests to the regional Mantle host with the Bedrock API key", async () => {
const model = mantleModels()[0];
const requests: { url: string; authorization: string | null }[] = [];