Jahanzaib

AI model leaderboard · checked 2026-10-10

Which AI model should you build on? Every score and price, sourced.

54 current models from 18 providers, priced from their own pages and scored by independent leaderboards. Click any model to see the source behind every number. No blanks filled with guesses, and no 2024 models padding the list.

Current models
54
Providers
18
Sourced measurements
841
Open weights
31

Price versus intelligence

The frontier: models nothing else beats on both price and score

Each dot is a model: further left is cheaper, higher is smarter, bigger means a longer context window. The dashed line joins the models that no other model beats on both price and intelligence. Anything below the line costs more for less.

0102030405060$0.1$1$10$100Blended price, USD per 1M tokens (3 input : 1 output, log scale)Intelligence Indexcheaper and smarter ↖Claude Opus 5.5Claude Sonnet 5.5Claude Haiku 5.5Ministral 3 8BMinistral 3 3BMuse Spark 1.3MiMo-V2.6-ProMiMo-V2.6-Flash

Best model for

The pick for each job, and the rule that picked it

Every pick comes from the data with the rule shown, so you can check it rather than trust it.

Leaderboards

Ranked by the benchmarks that still separate models

Old favourites like MMLU and HumanEval are saturated, so they are not here. Switch boards to see who moves.

A composite of 10 independent evaluations (agents 30%, coding 20%, scientific reasoning 20%, general 30%), run by Artificial Analysis on each model, version 4.3.2.

Source: Artificial Analysis
  1. 1
    57.6
  2. 2
    56
  3. 3
    53.4
  4. 4
    52.7
  5. 5
    51.8
  6. 6
    48.1
  7. 7
    46.5
  8. 8
    46.3
  9. 9
    45.4
  10. 10
    44.8
  11. 11
    43.6
  12. 12
    43.4
  13. 13
    41.8
  14. 14
    40.9
  15. 15
    39.9

What it costs you

Your monthly bill on each model, for your workload

Pick a workload or type your own volumes. Prices are each provider’s published list price, checked on the date above.

Choose models (12)

Standard list prices, cache reads at the provider’s cached input rate where one is published. Long context surcharges, batch discounts and reasoning tokens are not modelled, and newer Claude models count about 30% more tokens for the same text, per Anthropic.

Monthly bill for 20,000 requests, cheapest first

  1. 1
    $6.75
  2. 2
    $17.93
  3. 3
    $69.00
  4. 4
    $76.70
  5. 5
    $104
  6. 6
    $111
  7. 7
    $133
  8. 8
    $133
  9. 9
    $203
  10. 10
    $265
  11. 11
    $656
  12. 12
    $675

Every model

The full comparison

Sort by any column. Missing values stay at the bottom instead of pretending to be zero. Click a row for sources.

54 models

Method

How every number got here

  • Prices come from each provider’s own pricing page or model docs, checked on 2026-10-10. Where a provider prices only in another currency or not at all, the field says so instead of converting or estimating. Scheduled changes and promotions are quoted in the provider’s own words in each model’s details.
  • Blended price weights input three to one against output, a long standing convention for chat and agent workloads. Your mix may differ; the cost estimator uses your own token counts.
  • Intelligence and speed come from Artificial Analysis (Intelligence Index v4.3.2). Where a model is listed at several reasoning settings, the page uses its best scoring setting and the speed measured for that same setting.
  • Human preference comes from the Arena text leaderboard (style control on). Agentic coding, reasoning and expert knowledge come from the official Terminal-Bench, ARC Prize and Scale Labs leaderboards. Each entry keeps the exact name and setting the leaderboard prints.
  • Provider reported benchmarks appear only in each model’s details, labelled as such, because providers choose their own settings and harnesses.
  • Legacy models (superseded but still callable) are hidden by default. Gated models you cannot simply call are left out.

Questions

Choosing a model, answered

Which AI model is the best right now?+

On the Artificial Analysis Intelligence Index, Claude Opus 5.5 from Anthropic scores highest among current models (57.62). The best model for you depends on the job: agentic coding, reasoning, price and speed each have a different leader, which is why this page shows a pick per job with the rule behind it.

Which AI model API is the cheapest?+

Among models scoring 40 or more on the Intelligence Index, Claude Haiku 5.5 has the lowest blended price at $0.2 per million tokens. Smaller models cost less still, but check that they can do your task before switching; the cost estimator shows your real monthly bill.

What is the Artificial Analysis Intelligence Index?+

An independent composite score. Artificial Analysis runs ten evaluations on each model itself, weighted agents 30%, coding 20%, scientific reasoning 20% and general 30%, so models are compared on the same tests rather than on numbers each provider picked.

How is the blended price calculated?+

Three parts input price to one part output price, per million tokens. It is a convention for comparing models at a glance. Your own mix may differ, so the cost estimator uses your token counts and cache rate instead.

Are open weight models good enough for production?+

MiMo-V2.6-Pro is the highest scoring model whose weights you can download, at 46.32 on the index against 57.62 for the top closed model. Read the license in each model’s details: several add revenue thresholds or hosting clauses.

Why is SWE-bench Verified not on this page?+

Providers stopped reporting it for current models, and the official leaderboard has no current frontier entries. Agentic coding is ranked here with Terminal-Bench 4.0 from its official leaderboard instead.

Which model should I use for an AI agent?+

Rarely just one. Most agents I ship use a fast, cheap model for routine steps and a stronger model only for the steps that need it. Shortlist two or three models here, then test them on your own tasks, because benchmark rank and real accuracy on your data are not the same thing.

How often is this updated?+

Every price and score shows the date it was checked; this edition was checked on 2026-10-10. Prices and leaderboards move monthly, so treat anything older than a few weeks as a starting point and confirm on the provider’s page before you commit.

The leaderboard picks a model. Your workload picks the right one.

Claude Opus 5.5 tops the index today, but most agents I ship run on a cheaper model for most steps and a stronger one only where it pays. I can test your real tasks across the shortlist and give you the cost and accuracy side by side.

Book Meeting

AI agent cost calculator Build vs buy calculator