AI Model Leaderboard: LMArena Rankings, Prices & Releases

ESPN for AI. Get today's AI model standings from the LMArena leaderboard with week-over-week movement (rank change, score change, new entries). Every model comes matched to its API price per 1M tokens and context window from OpenRouter. The same run also returns new model releases and the trending open models on Hugging Face.

Covers every LMArena arena: Text (chat LLMs), WebDev (coding apps), Vision, Search, Document, Text-to-Image, Image Edit, Text-to-Video, Image-to-Video, Video Edit and Agent. It also covers categories such as coding, math, creative writing, hard prompts, instruction following, multi-turn, expert and 10+ languages.

What you get

• Standings: rank, model, lab, license, Arena score with confidence interval, votes, leaderboard date.
• Movement: previous rank, rankChange, scoreChange and movement (up/down/same/new) against the leaderboard from N days ago (default 7). LMArena sometimes renames models (e.g. gemini-3.5-flash-high → Gemini 3.5 Flash (High)); these are matched across snapshots and the old name is returned in previousName.
• Price and specs: OpenRouter model ID, input and output price per 1M tokens, and context length for each ranked model.
• New releases: models added to OpenRouter in the last N days with price, context, modalities, tool and reasoning support, and Arena rank when ranked.
• Trending: the hottest open-weight models on Hugging Face, with trending score, likes, downloads, task and license.

How to use

1. Pick arenas (default: Text) and categories (default: overall). For example, Text + coding gives the best coding LLMs.
2. Set Top N and how many days back to compare.
3. Click Start, or schedule it daily or weekly and send the results to Slack, Google Sheets, Notion or your newsletter tool.

An empty input returns the Text overall top 20 with weekly movement and prices, the last 14 days of releases, and the top 15 trending Hugging Face models.

Input example

{
  "sections": ["standings", "releases"],
  "arenas": ["text", "webdev"],
  "categories": ["overall", "coding"],
  "topN": 25,
  "compareDaysAgo": 7,
  "releasesWithinDays": 30
}

Output

Standing:

{
  "type": "standing", "arena": "text", "category": "overall", "rank": 1,
  "model": "Gemini 4 Argon (High)", "organization": "google", "license": "Proprietary",
  "score": 1525.7, "scoreLower": 1519.2, "scoreUpper": 1532.3, "votes": 10083,
  "previousRank": 1, "rankChange": 0, "scoreChange": -6.6, "movement": "same", "isNewEntry": false,
  "leaderboardDate": "2026-10-22", "comparedToDate": "2026-10-15",
  "openRouterId": "google/gemini-4-argon", "inputPricePer1M": 2, "outputPricePer1M": 10, "contextLength": 1048576,
  "source": "LMArena (lmarena.ai, CC BY 4.0)"
}

Release:

{
  "type": "release", "model": "MoonshotAI: Kimi K3.1 Pro", "id": "moonshotai/kimi-k3.1-pro", "provider": "moonshotai",
  "releasedAt": "2026-10-23T17:56:28.000Z", "contextLength": 1048576, "maxOutputTokens": 943718,
  "inputModalities": ["text", "image", "video"], "inputPricePer1M": 2, "outputPricePer1M": 10,
  "supportsTools": true, "supportsReasoning": true, "arenaTextRank": null, "url": "https://openrouter.ai/moonshotai/kimi-k3.1-pro"
}

Trending model:

{
  "type": "trendingModel", "rank": 1, "model": "deepseek-ai/DeepSeek-V4.1-Flash-1017", "pipelineTag": "image-text-to-text",
  "trendingScore": 2224, "likes": 2617, "downloads": 57715, "license": "mit", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
}

The Output tab has three views: Standings, New releases and Trending on Hugging Face.

How much does it cost?

Pay per event: $0.005 per run + $0.001 per row ($1 per 1,000). No compute charges.

You get	You pay (Starter/Bronze plan)
Default run (20 standings + up to 25 releases + 15 trending)	about $0.065
Daily default run for a month	about $1.95
4 arenas × 3 categories × top 50 (600 rows)	$0.605

Apify's free plan includes $5 of usage every month, and free-plan users pay half price. Paid plans get volume discounts (Silver -10%, Gold and above -20%).

Use cases

• AI newsletters, creators and media: a weekly "AI power rankings" with movers, new entries and prices, ready to publish.
• Product and engineering teams: pick the best model per dollar for coding, writing or vision. Re-check every week.
• Investors and analysts: follow which labs gain or lose ground, and how fast new models climb.
• Model routing and cost dashboards: feed rank, price and context into a router or a Sheets dashboard.
• AI agents: call it through the Apify MCP server to answer "what's the best coding model under $5 per 1M output tokens?".

Run it from code (API, Python, JavaScript)

curl -X POST "https://api.apify.com/v2/acts/api-co~ai-model-leaderboard/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" -d '{"sections": ["standings"], "categories": ["coding"], "topN": 10}'

from apify_client import ApifyClient
client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("api-co/ai-model-leaderboard").call(run_input={"arenas": ["text", "webdev"], "topN": 10})
for r in client.dataset(run["defaultDatasetId"]).iterate_items():
    if r["type"] == "standing":
        print(r["arena"], r["rank"], r["model"], r["movement"], r.get("outputPricePer1M"))

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: 'YOUR_APIFY_TOKEN' });
const run = await client.actor('api-co/ai-model-leaderboard').call({ sections: ['releases'], releasesWithinDays: 7 });
const { items } = await client.dataset(run.defaultDatasetId).listItems();

FAQ

Where does the leaderboard data come from? Can I republish it? From LMArena's official leaderboard dataset on Hugging Face (lmarena-ai/leaderboard-dataset). It is updated about daily and licensed CC BY 4.0, so you can republish it as long as you credit LMArena. Every standing row includes a source field for that. Earlier snapshots are read from the dataset's version history, and that history is how the rank changes are calculated.

How are prices matched? Model names are normalized (provider prefix and effort labels such as "(High)" or "(Max)" are removed) and matched to OpenRouter model names. When there is no confident match, openRouterId is null instead of a guess. This is common for image and video models that OpenRouter does not list. Prices are OpenRouter's list prices in USD per 1M tokens.

Why do some models appear twice, e.g. "(High)" and "(Max)"? LMArena ranks each reasoning-effort setting separately. Both rows map to the same API model and price.

What does "Text, style control" mean? A version of the Text leaderboard that adjusts for answer length and formatting, so models can't win just by writing longer or prettier answers.

How often does it update? LMArena publishes new standings about once a day (some arenas less often). OpenRouter and Hugging Face data are live. A daily schedule is enough.

What is the Agent arena score? The Agent arena uses a different scale (a small decimal win-rate style score), so compare it only within that arena.

Related Actors

• Agent Skills & MCP Server Directory: the most-installed agent skills and MCP servers.
• AI Bot Analytics & AI Crawler Checker: which AI bots visit your website.
• Academic Papers Search: find the papers behind the models.