Use Inference APIs with FluidVoice

Settings

Which model

meta-llama/Llama-3.3-70B-Instruct-Turbo for clean-up with no thinking delay. deepseek-ai/DeepSeek-V4.1-Flash is faster and cheaper if FluidVoice's reasoning setting is switched off for it. Prices and measured speed: chat models.

Steps

1. In FluidVoice open the AI settings and add a custom provider.
2. Base URL: https://api.inferenceapis.com/v1
3. API key: your Inference APIs key.
4. Let FluidVoice fetch the model list and choose meta-llama/Llama-3.3-70B-Instruct-Turbo. The list also contains speech and embedding models; pick a chat model.

Configuration

curl -s https://api.inferenceapis.com/v1/models -H "Authorization: Bearer $INFERENCE_API_KEY" | head -c 400

Verify

Dictate a sentence with AI enhancement on; the cleaned text should appear within about a second.

How this guide was checked

Configuration only We do not have a Mac to run FluidVoice on. FluidVoice transcribes on the device and uses a cloud model only for its AI enhancement and command modes. Its source shows custom providers take a base URL and key and then list models from /models; on 2026-09-18 we confirmed this API answers that call and the chat completions request that follows. If a step does not match what you see, tell us.

Gotchas

• Only text leaves the Mac with this setup. FluidVoice's transcription is local.
• Reasoning models add seconds of thinking to each dictation unless reasoning is off. Llama 3.3 70B has no reasoning step.
• A one-sentence clean-up is a few hundred tokens, well under a hundredth of a cent.

If something fails

401 — key missing or wrong · 402 insufficient_balance — add credits on Billing · 404 model_not_found — check the id against the model list (aliases such as gpt-oss-120b work too) · 503 model_unavailable — the model is being enabled. Full details in the API docs.