Skip to content
UtilPick.
My toolbox
All tools
Favorites
Saved tools

Loading…

Recently used

Tools you open will appear here.

Get the desktop app
Settings
UtilPick.
All tools
All tools/AI Model Leaderboard

AI Model Leaderboard

Compare AI models by category scores, value and speed from LiveBench and OpenRouter data, refreshed daily with the date shown, plus image and video rankings.

Loading the tool…

Used this tool? Share what worked for you — it helps the next person.

User reviews

No reviews yet.

Write a review

Checking your sign-in status…

    From first step to finished result

    AI Model Leaderboard: complete guide

    Walk through the steps and examples, or jump straight to the topic you need.

    How to use
    Examples & choices
    Troubleshooting
    FAQ

    At a glance

    Raw data
    The server fetches the LiveBench benchmark and OpenRouter model information once a day, recalculates the rankings, and shows the as-of date at the top of the screen.
    On days it cannot be fetched
    If the raw data cannot be fetched, the tool shows the rankings built into it (as of 2026-11-12) and tells you so.
    What is shown
    The track shows the top 12 models running, and the ranking table shows up to 30th place for each measure. Measures with fewer than 3 measured models are hidden.
    Measure groups
    There are groups for overall, by field, value, speed, reliability, and image, video and music.
    Price
    US dollars per million tokens, blended at 3 input to 1 output. The Korean won conversion is a reference value converted at the European Central Bank reference rate.

    How to use AI Model Leaderboard

    1. 1. Pick a benchmark tab and the companies you want to see

      01Pick a benchmark tab and the companies you want to see

    2. 2. Watch the top 12 models race to their places by score

      02Watch the top 12 models race to their places by score

    3. 3. See the top 30 in the table and click a model name to gather its ranks across every benchmark

      03See the top 30 in the table and click a model name to gather its ranks across every benchmark

    1. 01

      Choose what to compare

      In Comparison group, choose overall, by field, value and so on, then click a measure tab inside it. Each tab states what was measured and the direction, such as higher is better.

    2. 02

      Pick a company

      Choose a company under View by company and only that company's models remain. The rank numbers keep using the overall ranking.

    3. 03

      Read the track

      The top 12 models run in proportion to their values and line up in rank order. The actual values are written on the ticks above the track, so look at the ticks, not just the distances.

    4. 04

      Check the ranking table and model summary

      In the ranking table, click Show all to expand to 30th place, and click a model name to see its rank on every measure at once.

    What it’s good for

    It is split into overall, by field, value and speed

    Overall includes the LiveBench overall score and the intelligence, coding and agent indexes that OpenRouter provides (evaluated by Artificial Analysis). The LiveBench overall is a weighted mean of the seven LiveBench 2.0 suites (Long Agentic x2, Coding and Design x1.5, the rest x1). By field covers those seven suites: Long Agentic, Reasoning, Coding, Browser, Data Analysis, Research and Design. Value is the cost of solving one LiveBench problem and the price per million tokens, and speed and reliability are the output speed, time to first response, uptime and context length that OpenRouter measures. Being first on one measure does not mean being ahead on another, so look at the measure that fits what you want to do.

    How to read the value chart

    The horizontal axis is cost per success and the vertical axis is the score. The dotted blue line connects only the models for which no other model scores higher at the same cost. A model below or to the right of the line has an alternative that is cheaper and scores higher.

    Read the track ticks as well

    For measures where scores are close together, the track starts a little below the minimum rather than at 0 to make differences easier to see. Price and context length differ by tens of times, so they use a logarithmic scale. The actual values are written on the ticks above the track, so do not read distances directly as multiples.

    Refreshed once a day

    The raw data is fetched once a day, the rankings are recalculated, and the as-of date is shown at the top of the screen. Benchmarks change with the reasoning-effort setting (max, high and so on), so the value at the highest setting is shown. Price is in US dollars per million tokens, blended at 3 input to 1 output, and on the Korean-language screen a reference value in Korean won (marked with approximately) converted at that day's European Central Bank reference rate is shown beside it. Actual charges can differ by provider and discount terms.

    Worked examples

    Choosing the price tab

    Input and settings
    Comparison group: Value → Price
    Result
    The cheaper a model is, the closer it is to the finish line
    The track uses a logarithmic scale
    For price, lower is better. The values differ so much that the track is drawn on a logarithmic scale, so do not read distances as multiples.

    Reading the value line

    Input and settings
    Cost per success and score
    A $0.02, 60 points; B $0.05, 55 points
    C $0.10, 70 points; D $0.40, 68 points
    Result
    Value line: A → C
    Going from cheapest to most expensive, only models that score higher than every one so far are kept. B is more expensive than A and scores lower, and D is more expensive than C and scores lower, so both drop off the line.

    Viewing just one company

    Input and settings
    View by company: choose one company
    Result
    Only that company's models remain, and rank numbers stay as the overall ranking
    Even if a model is first within its company, its overall rank is shown too, so you can compare without exaggeration.

    Choosing a measure group

    Swipe the table sideways to see the rest.

    Choosing a measure group
    GroupWhat it showsSource
    OverallLiveBench overall (weighted mean of seven suites: Long Agentic x2, Coding and Design x1.5, the rest x1) and the intelligence, coding and agent indexesLiveBench, OpenRouter (evaluated by Artificial Analysis)
    By fieldLong Agentic, Reasoning, Coding, Browser, Data Analysis, Research, Design (the seven LiveBench 2.0 suites)LiveBench
    ValueCost per success and price per million tokensLiveBench, OpenRouter
    Speed and reliabilityOutput speed, first response, uptime, context lengthOpenRouter

    When the result isn’t what you expected

    It says built-in data

    The raw data could not be fetched that day, so the tool is showing the rankings built into it. Check the as-of date shown and open it again later.

    The model I am looking for is out of the ranking

    Only the top 30 are kept for each measure. If that measure was not measured for the model or it is below 30th, the summary says it is not in the top n or has no measured data.

    It says there are no models from the company I picked

    That company has no model in this measure's ranking. Try a different measure or company.

    Key terms

    LiveBench
    A benchmark that scores AI models on tasks across several fields. This tool uses the scores of the seven LiveBench 2.0 suites.
    Logarithmic scale
    A scale in which each tenfold increase in value moves the same distance. It is used for values that differ widely, such as price and context length.
    Value line
    A line connecting the models for which no other model scores higher at the same cost.

    AI Model Leaderboard FAQ

    What date are the rankings as of?

    The as-of date shown at the top of the screen. The server fetches the raw LiveBench and OpenRouter data once a day, so a new model is reflected within a day at most after it appears in the raw data. On a day when the raw data cannot be fetched, the tool shows the rankings built into it, along with their as-of date.

    What does being first on price mean?

    It is the cheapest model when 1 million tokens are used at a 3 input to 1 output ratio. On the price tab, the cheaper a model is, the closer it is to the finish line.

    Are the logos and characters official?

    No. The logos and characters are used only to refer to each company's trademark, and UtilPick is not affiliated with, sponsored by or endorsed by any of these companies.

    What appears when I click a model name?

    A summary window opens showing the model's rank on each measure. A measure where it is not in the top 30 is shown as out of the ranking.

    Sometimes the speed and first-response tabs are missing.

    A measure with fewer than 3 measured models is hidden because it is hard to call it a ranking. It appears again once the raw data fills back in.

    Jump to a topic

    • #AI model comparison
    • #measure groups
    • #value line
    • #built-in data

    Reviewed 2026.11.12

    Related tools

    View all
    Chart MakerTime & Work
    Character CounterText
    Unit Price ComparisonMoney & Math
    UtilPick

    Simple tools for a complicated day. A free collection of online utilities.

    Service

    About UtilPick Get the desktop app

    Help

    Request a toolReport a problemContactinfo@utilpick.com

    Policies

    Privacy Policy (Korean)Terms of Use (Korean)
    Business registration no. 365-47-00936 (Republic of Korea)
    © 2026 UtilPick
    Featured on Twelve ToolsUtilPick - Featured on Startup FameFazier badgeFeatured on LaunchIgniter
    HomeSearchFavoritesMore

    Today’s leaders

    • Today’s changes…
    01

    Model ranking race

    Fetching the latest rankings…

    LiveBench overall
    Intelligence index
    Coding index
    Agentic index
    Weighted average of LiveBench 2.0’s 7 category scores, using LiveBench’s own weights (currently Long Agentic ×2, Coding and Design ×1.5, the rest ×1)Higher is betterSource LiveBench

    LiveBench overall: #1 Claude Fable 5.2 · #1 overall · 84.9%

    Axis doesn’t start at 0
    Start 63.0%74.0%Finish 84.9%
    3
    Claude Fable 5.1Anthropic · 2026-09-02
    84.0%weighted avg. score
    1
    Claude Fable 5.2NEWAnthropic · 2026-11-10
    84.9%weighted avg. score
    12
    Claude Haiku 5.5Anthropic · 2026-10-08
    68.9%weighted avg. score
    5
    Claude Opus 5.5Anthropic · 2026-09-23
    83.4%weighted avg. score
    9
    Claude Sonnet 5.5Anthropic · 2026-09-29
    77.0%weighted avg. score
    11
    DeepSeek V4.1 Flash 1017NEWDeepSeek · 2026-10-17
    70.3%weighted avg. score
    8
    Gemini 4 ArgonNEWGoogle DeepMind · 2026-10-14
    77.7%weighted avg. score
    4
    Gemini 4 CarbonNEWGoogle DeepMind · 2026-11-11
    83.5%weighted avg. score
    2
    GPT-6 AstraOpenAI · 2026-09-05
    84.5%weighted avg. score
    6
    GPT-6.1 SolOpenAI · 2026-09-30
    81.4%weighted avg. score
    7
    GPT-6.1 TerraNEWOpenAI · 2026-10-27
    78.8%weighted avg. score
    10
    Muse WonderMeta AI · 2026-10-12
    70.7%weighted avg. score

    Today’s changes

    Daily rankings are collected and compared with yesterday

    Rankings started collecting today. From tomorrow, rises, drops and new entries compared with yesterday appear here.

    Value at a glance

    Value based on the real cost of solving LiveBench questions. Switch categories to compare.

    Models on the blue dashed line (value frontier) have no rival scoring higher for the same cost. Click a dot to see that model at a glance.
    Overall · cheapest cost per success first
    1. $0.13 · 60.4 pts
    2. $0.17 · 63.7 pts
    3. $0.17 · 60.7 pts
    4. $0.18 · 42.3 pts
    5. $0.18 · 66.1 pts
    6. $0.19 · 70.3 pts
    7. $0.22 · 32.4 pts
    8. $0.31 · 47.0 pts
    02

    Ranking table

    Weighted average of LiveBench 2.0’s 7 category scores, using LiveBench’s own weights (currently Long Agentic ×2, Coding and Design ×1.5, the rest ×1) · Higher is better · click a model name to see every benchmark

    Swipe the table sideways to see more columns.

    AI model ranking by LiveBench overall, as of 2026-11-12
    RankModelCompanyweighted avg. scoreReleased
    1 NEWAnthropic84.9%2026-11-10
    2OpenAI84.5%2026-09-05
    3Anthropic84.0%2026-09-02
    4 NEWGoogle DeepMind83.5%2026-11-11
    5Anthropic83.4%2026-09-23
    6OpenAI81.4%2026-09-30
    7 NEWOpenAI78.8%2026-10-27
    8 NEWGoogle DeepMind77.7%2026-10-14
    9Anthropic77.0%2026-09-29
    10Meta AI70.7%2026-10-12
    11 NEWDeepSeek70.3%2026-10-17
    12Anthropic68.9%2026-10-08
    Show sources and method
    As of
    2026-11-12 · source data is fetched and re-ranked once a day
    Source
    • LiveBench.AILiveBench 2.0 overall; long agentic, reasoning, coding, browser tasks, data analysis, research and design; cost per success, the value chart and the text AI replay
    • via OpenRouterPrice and context length, last-day uptime, last-30-minute output speed and time to first token, intelligence/coding/agentic indexes (Artificial Analysis)
    • Artificial AnalysisArena Elo scores for text to image, image editing, text to video, image to video, music with vocals and instrumental music
    Reading the track
    The top 12 models run on the track; see the rest with Show all in the table. Score benchmarks start a little below the lowest value so differences are visible. Price and context use a log scale. Don’t read distance as a multiple — check the tick values.
    Measurement conditions
    Benchmarks use each model’s highest reasoning effort (max, xhigh, high). Price is USD per million tokens, blended 3 input : 1 output. LiveBench overall averages the 7 category scores with the weights LiveBench sets for each release, and is left out if any category is missing. Cost per success is the same (cost per task ÷ score) × 100 that the LiveBench site uses. Release dates are when a model appeared on OpenRouter; models not on OpenRouter are left blank. NEW marks models released within 30 days of the reference date. Image, video and music scores are Elo ratings from blind side-by-side votes, so differences of a few dozen points are effectively a tie. Artificial Analysis does not report release dates, so these benchmarks show no release date or NEW mark.
    Trademarks
    Logos and characters refer to each company’s trademarks, which belong to their owners. UtilPick is not affiliated with, sponsored or endorsed by any of them. Logo icons: Lobe Icons (MIT).
    Other AI rankingsImage Generation AI RankingVideo Generation AI RankingAI Music Generator Ranking — Suno & More