Build guide

Build an AI model price tracker bot for X
price drops, new models, removals

New models land on OpenRouter several times a week and prices move without an announcement. Developers follow the accounts that notice first. One keyless API, one RSS feed and a 50-line script post it all, with no X developer account.

Card required, cancel any time.

The short answer

For new models, add https://openrouter.ai/api/v1/models?use_rss=true&sort=newest to the OpenTweet RSS connector with a one-line prompt. For price drops, fetch https://openrouter.ai/api/v1/models every hour, compare pricing.prompt and pricing.completion with the last run, and post each drop to POST https://opentweet.io/api/v1/posts. Skip -1 routers and :free and :batch variants, and multiply prices by a million.

This account type already runs on OpenTweet. In the last 14 days one paying API account posted more than 300 LLM price changes across model routers, about 20 a day, in the "Price drop: model, input old to new, output old to new" format this page builds. Other paying accounts post AI news and agent takes every day. Demand is there too: on Google Trends over the last three months, "openrouter" averaged about 30 times the searches for "twitter bot".

This week: GLM 5.1 leaves OpenRouter after November 20

On November 16, 2026, the models API listed 19 models with an expiration_date. The first is z-ai/glm-5.1 on November 20, followed by openai/gpt-5-image-mini on December 1 and openai/gpt-5, gpt-5-pro and gpt-5-image on December 11. Each one is a post that saves somebody a broken app. Qwen4 Max and two Gemini 4 models (Hydrogen and Carbon) were added in the same week.

Six posts a model price account runs on

The script covers drops, new models and removals. The RSS connector covers new models with no code. Scheduled posts cover the weekly roundups.

Price drop

When it happens

Old and new input and output price per million tokens, and the percentage. The script writes it.

New model

Several a week

Who made it, the price, the context window and what it accepts. The RSS connector or the script.

Removal warning

A few a month

A model gets an expiration_date. Developers who still call it need to know.

Free model alert

As listed

A new :free variant, with OpenRouter's free limits: 20 requests a minute, 50 a day under $10 of credits.

Cheapest by job

Weekly

The cheapest model with tools, with images, with 1M context this week. Scheduled, from the same snapshot.

Launch thread

Big days

A big release (a new Claude, GPT or Gemini) gets a thread: price vs the last version, context, where to try it.

No code: new models from OpenRouter's RSS feed

OpenRouter's models endpoint turns into an RSS 2.0 feed with use_rss=true. We requested each URL below with the OpenTweet connector's own User-Agent on November 16, 2026; all returned 200. Each item is one model with its name, ID, a short description and a pubDate.

BeatFeed URLItems
Every new modelopenrouter.ai/api/v1/models?use_rss=true&sort=newest468
Models that accept tools (agents)...?use_rss=true&sort=newest&supported_parameters=tools380
Image models...?use_rss=true&sort=newest&output_modalities=image72
Coding models...?use_rss=true&sort=newest&category=programming20

The feed returns the whole catalog every time (limit is ignored in RSS mode), so the item counts are large. That is fine: scheduled connector runs skip items older than 48 hours and post at most the number you set per run, so only new models get posted. Add use_rss_chat_links=true to make each item link to the OpenRouter chat with that model.

Set it up in the RSS connector

Connectors  ->  New connector  ->  Blog or newsletter (RSS)

  Feed URL:   https://openrouter.ai/api/v1/models?use_rss=true&sort=newest
  Prompt:     One post for developers about this new model: who
              made it, what it is for, and one standout spec
              (context window, inputs, reasoning). Plain words,
              no hype, no "game changer". Credit OpenRouter.
              Skip items whose ID ends in :batch.
  Frequency:  Every 15 minutes
  X account:  @yourmodelwatch

Batch variants (:batch, 86 of them on November 16) are listed as separate models, which is why the prompt skips them. The feed carries no prices; for those, use the script.

Price drops, new models and removals in Python

Fetches the whole catalog, compares it with the last snapshot, and posts each price drop, each new paid model, and each new removal date. The first run only saves the snapshot. Run it from cron every hour. With DRY_RUN=1 it prints instead of posting.

import os, json, pathlib, requests

DRY = os.environ.get("DRY_RUN") == "1"
OT = {"Authorization": f"Bearer {os.environ.get('OPENTWEET_API_KEY', '')}"}
SNAP = pathlib.Path("openrouter_models.json")

def per_million(p):  # OpenRouter prices are USD per token, as strings
    return float(p) * 1_000_000

def fmt(x):
    return f"${x:,.2f}" if x >= 0.1 else f"${x:,.3f}"

MAX_POSTS, sent = 10, 0  # a big release day can change dozens of prices

def post(text):
    global sent
    if sent >= MAX_POSTS:
        return
    sent += 1
    if DRY:
        print("---\n" + text); return
    r = requests.post("https://opentweet.io/api/v1/posts", headers=OT,
                      json={"text": text, "publish_now": True}, timeout=20)
    if not r.ok:
        print("OpenTweet error", r.status_code, r.text[:200])

models = requests.get("https://openrouter.ai/api/v1/models", timeout=30).json()["data"]
now = {m["id"]: m for m in models
       if m["pricing"].get("prompt") not in (None, "-1") and not m["id"].endswith((":free", ":batch"))}
old = json.loads(SNAP.read_text()) if SNAP.exists() else None

if old is not None:
    for mid, m in now.items():
        was = old.get(mid)
        p_in, p_out = per_million(m["pricing"]["prompt"]), per_million(m["pricing"]["completion"])
        if was is None:
            if p_in > 0:
                post(f"New on OpenRouter: {m['name']}\n\n"
                     f"{fmt(p_in)} in / {fmt(p_out)} out per 1M tokens, "
                     f"{m['context_length']:,} context\n\nModel ID: {mid}")
            continue
        w_in, w_out = per_million(was["prompt"]), per_million(was["completion"])
        cheaper = p_in < w_in * 0.95 or p_out < w_out * 0.95  # ignore rounding noise
        if cheaper and p_in <= w_in and p_out <= w_out:
            cut = round(100 * (1 - (p_in + p_out) / (w_in + w_out)))
            post(f"Price drop: {m['name']} ({cut}% cheaper)\n\n"
                 f"Input {fmt(w_in)} -> {fmt(p_in)}\nOutput {fmt(w_out)} -> {fmt(p_out)}\n"
                 f"per 1M tokens\n\nModel ID: {mid}")
        if m.get("expiration_date") and not was.get("expiration_date"):
            post(f"Heads up: {m['name']} ({mid}) is scheduled to leave OpenRouter "
                 f"after {m['expiration_date']}. Move your traffic before then.")

SNAP.write_text(json.dumps({mid: {**m["pricing"], "expiration_date": m.get("expiration_date")}
                            for mid, m in now.items()}))

How we tested it, on November 16, 2026, against the live API: the first run saved 468 models minus routers, free and batch variants, and posted nothing. The second run posted nothing. Then we edited the snapshot to fake yesterday: Qwen4 Max at twice its price, Gemini 4 Hydrogen missing, GLM 5.1 with no removal date, and Gemini 4 Carbon cheaper on input but dearer on output. The run wrote the three posts below and skipped Carbon, because a mixed change is not a price drop. With a bad key, OpenTweet's 401 is printed and the run carries on.

What it posted

Price drop: Qwen: Qwen4 Max (50% cheaper)

Input $3.00 -> $1.50
Output $9.00 -> $4.50
per 1M tokens

Model ID: qwen/qwen4-max
Heads up: Z.ai: GLM 5.1 (z-ai/glm-5.1) is scheduled to leave OpenRouter after 2026-11-20. Move your traffic before then.

The posts name the model ID instead of a URL, so they are plain-text posts and don't use a link post. To add openrouter.ai/qwen/qwen4-max, count it as a link post (see the plan table). The cap of 10 posts a run keeps a big release day from flooding the account. Changes past the cap are skipped, not queued, because the snapshot is saved at the end, so set MAX_POSTS to what your plan allows.

Three values to filter out. "-1" means a router (openrouter/auto, typesafe/jev-router, openrouter/fusion and five more on November 16); its cost depends on the model it picks. "0" means free. :batch IDs are the same model at batch prices. Leave these in and your "cheapest model" post is wrong.

The endpoint is free, but be polite. OpenRouter lists auth for the models endpoint as optional and publishes no rate limit for it. Hourly is plenty; prices do not change by the minute.

Let an AI agent write the launch threads

Numbers come from the script; context comes from a model. Connect Claude, ChatGPT, Cursor or any OpenRouter model to the OpenTweet MCP server and ask: "Here are this week's new models and price changes from my snapshot. Write a five-post thread comparing them on price per million tokens and context, schedule it for Friday 9 a.m. ET, and save it as a draft." Running the agent on OpenRouter itself? See post to X from OpenRouter, and Jev Router for letting the router choose the cheapest model that can do the job.

Which plan fits your account

Flat monthly price, no per-post fee. The deciding numbers are posts a day and how many carry a link.

SetupTypical volumePlan
New models and price drops as plain-text posts, one accountUp to 20 posts a dayPro, $19.99/mo
Every price change across routers (about 20 a day), with model page links, or separate coding, image and free-model accountsUp to 100 posts a day, 10 link posts, 3 X accountsAdvanced, $29/mo
A network: one account per lab or per languageUp to 300 posts a day, 20 link posts, 10 X accountsAgency, $49/mo

The account we described above posts about 20 a day, which is Pro's whole daily limit, so a tracker that also posts news should start on Advanced. Pro includes 1 link post a day; link credits cover busy launch days on any plan. See pricing.

Prices are facts; descriptions are someone's writing. Post the numbers and a sentence of your own. Don't paste the model description, and credit OpenRouter as the source.

Check before you call it a cut. A price can drop because a cheaper provider joined, then rise when it leaves. Saying "on OpenRouter" and the date keeps the post true.

Label the bot. Turn on X's Automated label and don't use a lab's logo or name as your account name. More in X automation rules.

No affiliate spam. If you add referral links, say so. Accounts that post the same link over and over get limited. See the affiliate deals bot for what works.

Frequently asked questions

How do I track AI model price changes automatically?

Fetch https://openrouter.ai/api/v1/models on a schedule (it answered without an API key on November 16, 2026), save each model's pricing.prompt and pricing.completion, and compare with the last run. Any price that fell is a post. The script on this page does that and publishes each change to X through OpenTweet, with no X developer account.

Does OpenRouter have an RSS feed for new models?

Yes. Add use_rss=true to the models endpoint: https://openrouter.ai/api/v1/models?use_rss=true&sort=newest returns an RSS 2.0 feed titled "OpenRouter - New Models", one item per model with a pubDate. Filters work too: supported_parameters=tools, output_modalities=image, category=programming. The RSS feed has no prices, so use the API for price drops.

What unit are OpenRouter prices in?

US dollars per token, as strings. "0.0000015" is $1.50 per million tokens. Multiply by 1,000,000 before you post. A price of "-1" marks a router such as openrouter/auto or typesafe/jev-router, whose cost depends on the model it picks, and "0" marks free variants, usually IDs ending in :free.

How many models are on OpenRouter?

The models API listed 468 on November 16, 2026: 405 accept tools, 14 are :free variants, 86 are :batch variants at batch prices, and 38 were added in the previous 30 days. The newest was Qwen4 Max (qwen/qwen4-max), added that morning at $1.50 input and $4.50 output per million tokens.

How do I know when a model is being removed from OpenRouter?

Watch the expiration_date field. OpenRouter documents it as the date after which the model may be removed. On November 16, 2026, 19 models had one, the earliest z-ai/glm-5.1 on November 20, followed by several OpenAI GPT-5 models in December. A "heads up, move your traffic" post is useful to developers and rarely posted by anyone else.

Is it OK to post model prices from OpenRouter on X?

Prices and model names are facts, and OpenRouter publishes them in a public endpoint. Write the post yourself, credit OpenRouter as the source, and don't copy whole model descriptions. Don't present the account as OpenRouter or a model lab, and turn on X's Automated label for a bot.

More AI beats: the AI news account, the GitHub trending bot and the AI video account. Prices of a different kind? See the stock market bot.

Everything you can build on X with OpenTweet

Post to X from your code, a watcher, or an AI agent. Start anywhere in the cluster.

OpenTweet vs the X API

Post, schedule, and automate X without a developer account or the $200/mo minimum.

X DM outreach, human-approved

Find and qualify leads, AI-draft DMs, approve each one, and drip-send from your own account via API or MCP.

Post to X without an API

The clean, account-safe way to post to X from your code or an AI agent.

Twitter MCP Server

Give Claude, Cursor, and OpenClaw the ability to post to X. 44 tools included.

Developer docs

Quickstart, API keys, MCP setup, and the REST reference for posting to X from code or an agent.

Cross-posting reference

One call posts to X, Bluesky and LinkedIn. The platforms field, per-network results, and why a network gets skipped.

Social media API for AI agents

Three ways an agent can post to X, Bluesky and LinkedIn, and the trade-offs of each.

LinkedIn MCP server

Post and schedule to a LinkedIn profile from Claude or Cursor through the official LinkedIn API.

XMCP vs OpenTweet

X's official MCP server bills per API call and cannot schedule. Compare it with the flat-fee hosted MCP.

MCP for AI agents

Connect your AI client to X in under two minutes, no X developer account.

Developer API and keys

REST endpoints, one bearer key, and usage tracking. Build on OpenTweet.

OpenTweet for AI agents

The posting layer for autonomous agents and automations that live on X.

Build an AI Twitter persona

Give your AI agent its own X account. Setup, cadence, and the rules that keep it safe.

Best MCP servers for social media

The 2026 ranked list. How the hosted OpenTweet MCP compares with the official X MCP and others.

Best API to post to X

Seven options priced at real volumes, and why links decide the bill.

Cheapest way to post to X via code

Flat fee vs pay-per-use. Why the $0.20 per-link fee flips the math past ~145 link posts a month.

X API pricing 2026

Every pay-per-use X API endpoint price from docs.x.com in one searchable table.

Bluesky posting API

Post and schedule to Bluesky with one REST call and OAuth, and cross-post to X and LinkedIn in the same request.

Automate LinkedIn posts

Post to LinkedIn from an AI agent, n8n or a script through the official Posts API, with one REST or MCP call.

Be the account that posts the price drop first

Connect X, add the OpenRouter feed, and run the script before the next model launch.