DeepSeek V4 Flash was retired: a free, OpenAI-compatible replacement

Update, 2026-09-24: confirmed at the provider. NVIDIA now answers 410 Gone for deepseek-ai/deepseek-v4-flash-0731: "reached its end of life on 2026-09-21T08:00:00Z". InferAll therefore returns 400 model_retired for that id, immediately and with no fallback, so your code fails fast instead of waiting. A newer DeepSeek V4.1 Flash is listed by NVIDIA but timed out on every test (30s, 45s, 60s), so we do not serve it either. We serve no DeepSeek model today.

Update, 2026-09-21: NVIDIA delisted this model, so it is retired. It was the only DeepSeek id we had, and no other spelling reaches DeepSeek V4 through InferAll. Between 09-21 and 09-24 a request for it was answered by another free model; since 2026-09-24 it returns 400 model_retired instead (see above), so check your error handling rather than the response's model field.

The snippets below have been changed to nvidia/nemotron-3-ultra-550b-a55b, which is $0 in/out on the same free tier and routes as written, so they run correctly as printed. The DeepSeek id is kept in this note only to explain what changed. Check GET /ai/v1/models for what is live before trusting any roster in an article, including this one.

DeepSeek V4 is one of the strongest open-weight model families for reasoning, coding, and agentic work. InferAll served it free via NVIDIA NIM as deepseek-ai/deepseek-v4-flash-0731 until NVIDIA delisted it (see the update above), and it is not callable today. What the rest of this page describes still holds for the free tier generally: no credit card, $0 within the free tier, and it works with the OpenAI SDK you already have. The runnable examples use nvidia/nemotron-3-ultra-550b-a55b, a $0 model that routes as written.

Model	Status	Best for
deepseek-ai/deepseek-v4-flash-0731	Retired: NVIDIA delisted it, 2026-09	was: cost-efficient, lower-latency, high-volume
nvidia/nemotron-3.5-lightning-30b-a3b	Live, $0	Cost-efficient, lower-latency, high-volume
nvidia/nemotron-3-ultra-550b-a55b	Live, $0	The largest free model we serve: see the Ultra guide

The replacement is $0 on the free NVIDIA NIM tier.

---

Quick start (Python)

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, no card required
)

# Fast, cost-efficient, great default
response = client.chat.completions.create(
    model="nvidia/nemotron-3-ultra-550b-a55b",
    messages=[{"role": "user", "content": "Refactor this function for readability: ..."}],
    max_tokens=512,
)

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

The only change from calling OpenAI directly is the base_url. Your existing code, streaming, and tool use all work unchanged.

---

TypeScript

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: "Write a SQL migration to add a nullable column." }],
  max_tokens: 400,
});

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

---

Why call the replacement through InferAll

It's $0 in / $0 out. Nemotron runs on NVIDIA NIM, which InferAll exposes at $0 on the free tier.

One key for everything. The same ifu_... key also calls the other free open models, plus paid OpenAI, Anthropic and Gemini models when you need them.

OpenAI-compatible. Standard chat.completion responses, streaming, tool use and JSON mode, with whatever OpenAI client you already have. Only the base_url and the model change.

---

If you specifically need DeepSeek

We don't serve it today, and we won't route a DeepSeek request to another model under DeepSeek's name. If NVIDIA restores a DeepSeek model that answers reliably, it will appear in GET /ai/v1/models and this page will say so. If you came here for a strong free open model, the closest we have is NVIDIA Nemotron 3 Ultra 550B, measured and verified on 2026-09-24.

---

Compare it yourself

The best way to pick a model is to watch several answer the same prompt. The full free roster is one call away: curl https://api.inferall.ai/ai/v1/models, so you never hardcode a list that goes stale.

Sign up at inferall.ai/keys, then activate via the $5 starter pack at /billing. 25 free NIM calls come before any payment; the $5 becomes spendable balance for premium providers and unlocks ongoing access to the 10+ NIM open models at $0 in/out (within the free-plan daily request caps).