Skip to content

7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.

TypeScript
2.3 KB
Raw Blame

Typhoon

Typhoon uses https://api.opentyphoon.ai/v1 with bearer authentication for OpenAI-compatible chat completions (including streaming) and audio transcription. Create a key at Typhoon Playground, add it on the Keys page, or import TYPHOON_API_KEY / OPENTYPHOON_API_KEY. Auth JSON recognizes typhoon, opentyphoon, and open-typhoon.

The hosted API is an ongoing free research showcase, not a monthly credit grant. It is rate limited, provided without a production SLA, and collects usage data to improve its models and API. Do not send sensitive content without reviewing its data-use terms. The provider recommends its paid infrastructure partner for production-grade availability.

This integration covers chat and transcription only. Live-tested on 2026-10-07:

  • typhoon-v2.5-30b-a3b-instruct: chat, SSE streaming, and tool calling.
  • typhoon-asr-realtime: Thai speech transcription.
  • typhoon-isan-asr-realtime: Thai / Isan speech transcription (smoke-tested with a standard Thai sample, not an Isan accuracy benchmark).

OCR is intentionally excluded. The older typhoon-v2.1-12b-instruct still appears in documentation but returned 400: Model not found during testing.

Model rows are published only through the signed hosted catalog. No migration, bundled seed, or automatic model discovery bypasses the existing Premium-now / Free-after-30-days gate. Older app versions must update to recognize Typhoon.

Key validation deliberately requests a nonexistent model against the chat endpoint: the public /models endpoint cannot validate a key. Live controls confirmed the exact 400 {"detail":"Model not found"} response follows successful authentication; invalid/missing credentials return 401/403. Other responses, including rate limits and outages, remain inconclusive.

Transcription sends only multipart file and model, and consumes the upstream JSON text field. The existing route renders JSON/plain text locally. Whisper decoding options are not forwarded, and no native VTT/SRT or timestamps are advertised. The existing global upload ceiling is unchanged.

Sources: FAQ, authentication, ASR, rate limits.