Meta Llama 4 Maverick free API: retired, and the free models to use instead

Correction, 2026-09-02: pricing. An earlier version of this post said premium provider tokens are billed at the provider's rate with zero markup. That was not accurate: InferAll bills premium tokens at its own published per-token rates, which sit above the underlying provider cost. Current rates are on /pricing. The post has been corrected; nothing else about the models or the setup has changed.

Update, 2026-07-27: meta/llama-4-maverick-17b-128e-instruct is no longer served. No Llama 4 Maverick route is in the current InferAll catalog. This post is kept because the architecture discussion below is still useful and people search for it, but every runnable example here uses nvidia/nemotron-3-ultra-550b-a55b, a current $0 model whose production hot-route probe reports HTTP 200. The full live list is one request away: curl https://api.inferall.ai/ai/v1/models.

The original Llama 4 Maverick route is unavailable through InferAll today. The examples below show the same OpenAI-compatible call shape against a current $0 NVIDIA NIM model.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.inferall.ai/v1",
    api_key="ifu_your_key_here",  # get one at inferall.ai/keys
)

response = client.chat.completions.create(
    model="nvidia/nemotron-3-ultra-550b-a55b",
    messages=[{"role": "user", "content": "What makes a mixture-of-experts model different from a dense one?"}],
    max_tokens=512,
)

print(response.choices[0].message.content)

---

What is Llama 4 Maverick?

Llama 4 Maverick is Meta's 17B active parameter Mixture of Experts (MoE) model with 128 experts (17b-128e). The MoE architecture activates a subset of its 128 expert networks per token, giving it performance significantly above its active parameter count while keeping inference costs low.

Maverick sits in the Llama 4 family alongside Llama 4 Scout (smaller, faster) and is optimized for instruction-following, reasoning, and code. When it was listed, InferAll routed it through NVIDIA NIM at zero cost; today you should choose a current free ID from the live catalog.

---

TypeScript / Node.js

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.inferall.ai/v1",
  apiKey: process.env.INFERALL_API_KEY,
});

const response = await client.chat.completions.create({
  model: "nvidia/nemotron-3-ultra-550b-a55b",
  messages: [{ role: "user", content: "Explain mixture of experts architectures." }],
});

console.log(response.choices[0].message.content);

Streaming

with client.chat.completions.create(
    model="nvidia/nemotron-3-ultra-550b-a55b",
    messages=[{"role": "user", "content": "Walk me through mixture-of-experts architectures."}],
    stream=True,
) as stream:
    for chunk in stream:
        print(chunk.choices[0].delta.content or "", end="")

Claude Code / Cline / Cursor

Point any Anthropic-compatible client at InferAll and pin a current free model, or use an anthropic/ prefix when you want real Claude:

export ANTHROPIC_BASE_URL=https://api.inferall.ai
export ANTHROPIC_API_KEY=ifu_your_key_here
export ANTHROPIC_MODEL=nvidia/nemotron-3-ultra-550b-a55b

---

Free models on InferAll (selection)

Model	Size	Notes
nvidia/nemotron-3-ultra-550b-a55b	550B (55B active)	Current hot-route-probed free default
openai/gpt-oss-20b	20B	OpenAI's open-weight model, fast general tasks
z-ai/glm-5.3-flash	31B	General-purpose open model

The full free model list is always available at the API. Filter by inputPerM: 0.

---

Get started

inferall.ai/keys — sign up free and start calling (25 free NIM calls before any payment; the $5 starter pack unlocks ongoing use) at /billing. The $5 becomes spendable balance: 10+ open NIM models stay $0 in/out against it (within the free-plan daily request caps); premium providers (OpenAI, Anthropic, Google) bill at our published per-token rates.