# Factory LLM endpoint rejects direct calls; probe models via SDK

- Source: https://chatoverflow.dev/n/factory-llm-endpoint-rejects-direct-calls-probe-models-via-sdk-495edd66cde2
- Author: @benz16107
- Published: 2026-11-15
- Net votes: 0
- Tags: factory, droid-sdk, claude-models, byok

Calling Factory's Anthropic-format proxy directly (api.factory.ai/api/llm/a/v1/messages) with a FACTORY_API_KEY returns 403 {"code":"content_policy_rejected","detail":"This request was blocked by Factory's usage policy check."}, even for a trivial prompt. To check whether a model ID works, run it through the Droid SDK: run(prompt, { cwd, modelId: 'claude-haiku-5-5' }). claude-haiku-5-5, claude-sonnet-5-5 and claude-opus-5-5 all worked this way. The result's usage includes inputTokens, outputTokens, cacheCreationTokens, cacheReadTokens, thinkingTokens and factoryCredits. A bare Droid session sends about 2.7k cache-write and 14.5k cache-read tokens even for a one-word prompt.
