AI Hub Phase 8: Adding DeepInfra and Liquid AI — Now at 33 Providers

#flutter #supabase #ai #buildinpublic

AI Hub Phase 8: Adding DeepInfra and Liquid AI — Now at 33 Providers

What's New in Phase 8

Two more inference platforms join the hub, bringing the total to 33 providers:

DeepInfra

Endpoint: https://api.deepinfra.com/v1/openai/chat/completions
Default model: meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo
Why: One of the most cost-effective inference providers. Hosts hundreds of open models with an OpenAI-compat endpoint.

Liquid AI

Endpoint: https://api.liquid.ai/v1/chat/completions
Default model: liquid/lfm-40b
Why: Liquid Foundation Models (LFMs) use a novel architecture beyond transformers. Worth experimenting with for long-context tasks.

Same 7-Line Pattern

deepinfra: {
  displayName: "DeepInfra",
  envKey: "DEEPINFRA_API_KEY",
  chatUrl: "https://api.deepinfra.com/v1/openai/chat/completions",
  defaultModel: "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
  buildBody: OPENAI_COMPAT_BODY,
  parseResponse: OPENAI_COMPAT_PARSE,
},
liquid: {
  displayName: "Liquid AI",
  envKey: "LIQUID_API_KEY",
  chatUrl: "https://api.liquid.ai/v1/chat/completions",
  defaultModel: "liquid/lfm-40b",
  buildBody: OPENAI_COMPAT_BODY,
  parseResponse: OPENAI_COMPAT_PARSE,
},

The Pattern That Scales

Every provider using OpenAI-compat format costs 7 lines. No custom auth, no custom parsing. The OPENAI_COMPAT_BODY + OPENAI_COMPAT_PARSE pair handles everything.

Provider Count: 8 Phases In

Phase	Count	Focus
1	14	Major LLMs (OpenAI, Anthropic, Google, xAI, DeepSeek)
5	23	Chinese providers (Moonshot, Qwen, Zhipu, 01.AI)
6	29	Enterprise (Reka, Writer, MiniMax, Meta Llama, Nebius)
7	31	Platforms (Replicate, Coze)
8	33	Inference infra (DeepInfra, Liquid AI)