New free LLM API credits added in mid-November 2026

Free LLM API offers appear, shrink and disappear quickly, which is why a stale round-up is often worse than no list at all. This article covers providers added to the Free LLM API Tokens catalogue between 13 and 16 November 2026. The figures come from that tracker's snapshot at 16:00 UTC on 16 November 2026, when the list held 126 entries. I will not invent a daily cap. Where a provider does not publish tokens per day, I will say so. One-time grants are not a renewable free tier, and an OpenAI-compatible base_url is not a promise that every SDK feature works.

Official sign-up credits

Baichuan AI was added on 16 November. The documented base URL is https://api.baichuan-ai.com/v1. New accounts receive a one-time ¥80 sign-up credit, described as about 10 million Baichuan2-Turbo tokens, valid for three months. No card is required. Tokens per day are not published, and rate limits were not specified on the pages recorded by the tracker. The console is Chinese-language, on a China-based platform.

Baidu Qianfan was added the same day. The documented base URL is https://qianfan.baidubce.com/v2. The offer is one million free tokens on each of 17 models — the tracker names ERNIE 4.5, DeepSeek, Qwen3 and Kimi K2 — also one-time and valid for three months. The credit is for online inference only, not batch jobs. Tokens per day are not published. Whether a card is required was not specified. This is also a China-based platform with a Chinese-language console.

AMD Token Lab was added on 14 November. The documented base URL is https://api.tokenlab.amd.com/v1. Free members of the AMD AI Developer Program receive complimentary Token Lab points. The tracker records 10 points and 30-day access, as described in the AMD developer community, extendable once by 15 days, at about 20 requests per minute per key. Points are drawn at a per-model rate. No card is required, and tokens per day are not published. The tracker lists open models on AMD GPUs — DeepSeek V4, Qwen3.8, Gemma 4 and GLM-5.3 — but AMD does not publish a point amount on the AI Developer Program page, so confirm the "10 points" figure before you rely on it.

One-time grants on gateways

Two further additions are gateways, not first-party model hosts, and the tracker flags both for relay risk. A compatible SDK call looks the same as it does against an official API, but the operator, the logs and the privacy policy still belong to the service behind the base URL. Do not send private source code or personal data through a flagged relay if that is a concern.

GonkaRouter was added on 13 November. The documented base URL is https://api.gonkarouter.io/v1. New accounts that sign in with email or Google receive 1 million free tokens as a one-time grant, not a daily quota. No card is required, and tokens per day are not published. The tracker lists GLM-5.3-Flash, DeepSeek-V4-Flash and MiniMax-M2.7, and notes an Anthropic-compatible endpoint as well. A blog post mentions a $20 credit campaign that the pricing page does not confirm, so the figure to trust for now is the 1 million one-time tokens.

Vectoree was added the same day. The documented base URL is https://vectoree.ai/api/v1. After a CLI login and project link, new accounts receive a $5 one-time welcome credit, then a pay-as-you-go wallet. No card is required for that credit. Tokens per day are not published. The tracker describes an OpenAI-compatible model gateway with 36+ models, including vectoree/auto. The credit is one-time, not daily, and an empty wallet stops the gateway.

Other additions in the same window

Official additions in the same window included Braintrust's $0 Starter plan, with $10 of built-in model credits each month that do not roll over; Liquid AI's limited-time $2 credit for new accounts; Together AI listing Ternary Bonsai 27B as free for input and output; Novita AI listing Apodex 1.1 Mini as free for input and output; and IBM watsonx.ai's Free plan of up to 300,000 foundation-model tokens per month (a simple OpenAI-compatible base URL was not clearly published on the pages checked). NLP Cloud has a free plan to test its I models without a card, but it is not OpenAI-compatible.

Relay-flagged gateways added alongside them include NFlashX ($1 welcome credit for the first 100 accounts, plus a Flash model at 2 requests per minute and 4K output), Qora API (a $5 trial credit only on request, not automatic), and Speka ($1 of model usage each month, 10 requests per minute, one API key). Together and Novita mark a single model as free on a page that also lists paid models — re-check those rows before you depend on them.

A short OpenAI-compatible example

If the provider documents an OpenAI-compatible endpoint, the official Python client usually needs a published base URL, a key in the environment, and a model id copied from that provider's docs. Here is that pattern pointed at Baichuan's published base URL. Replace MODEL_ID with the id from its documentation.

import os from openai import OpenAI client = OpenAI( api_key=os.environ["API_KEY"], base_url="https://api.baichuan-ai.com/v1", ) resp = client.chat.completions.create( model="MODEL_ID", messages=[{"role": "user", "content": "Reply with the word ok."}], ) print(resp.choices[0].message.content) 

To point this at AMD Token Lab instead, set base_url to https://api.tokenlab.amd.com/v1 and copy a model id from AMD's docs. If a published base URL already ends in /v1 or /openai/v1, copy it as published and do not add a second /v1.

What not to assume

A three-month sign-up credit will not refill on the first of the month. A monthly credit that does not roll over is not a bank of unused tokens. "Tokens per day not published" means the public pages did not give a number, not that usage is unlimited.

The living list, and a public data.json file, is at https://freellmapi.link/. Mid-November additions are grouped at https://freellmapi.link/new. I am not affiliated with the providers. If an official page disagrees with a row, trust the official page.