Skip to main content
Van Data Team

Data verified November 6, 2026

AI model benchmarks

The newest AI models ranked on independent benchmarks, price and context window. Model records include linked official sources, and each score names the leaderboard it comes from.

Download JSON or CSVMethodology
Latest release
GPT-6 CyberOpenAI · Nov 5
New in the last 30 days
23models
Top Arena Text score
1,528Gemini 4 Carbon
Current models
125from 30 providers

Share this

Leaderboard

The newest models, ranked

Newest release first. Narrow it to this week or this month, rank it by any benchmark, price or context window, or keep one provider. Scores come only from independent leaderboards; # is the model's rank among current models on that benchmark.

Rank by
Provider

73 language models

Current language models with their independent benchmark scores, prices and release dates
#ModelArena TextElo
1
GPT-6 CyberNew

OpenAI · limited access · Nov 5, 2026

Not measured
2
Inkling 2 SmallNew

Thinking Machines Lab · Nov 5, 2026

Not measured
3
Hy4New

Tencent · Nov 4, 2026

Not measured
4
LongCat-2.5New

Meituan · Nov 2, 2026

Not measured
5
GPT-Rosalind-6New

OpenAI · limited access · Oct 30, 2026

Not measured
6
Gemini 4 CarbonNew

Google · limited access · Oct 28, 2026

1,528#1
7
GPT-6.1 TerraNew

OpenAI · Oct 26, 2026

Not measured
8
Kimi K3.1 ProNew

Moonshot Kimi · open weights · Oct 23, 2026

Not measured
9
Mercury 3New

Inception · Oct 19, 2026

Not measured
10
MiniMax M3.1 FlashNew

MiniMax · open weights · Oct 19, 2026

Not measured
Scores fromArena · 2026-11-03Epoch AI · 2026-11-03

Charts

The market at a glance

Three views of the current language models on Arena Text, the independent benchmark that covers the most of them. Hover any point to name it.

Quality against price

Arena Text · measured Oct 28, 2026

5 current models sit on the price and quality frontier: none is beaten on both. The cheapest is MiMo-V2.6-Flash at $0.18 per 1M tokens blended, the strongest Gemini 4 Carbon at 1,528.

Quality against priceArena Text score against blended token price for 44 current language models.1,2501,3001,3501,4001,4501,5001,550$0.30$1$3$10MOST ATTRACTIVE QUADRANTMedian price $3Median 1,476Blended price per 1M tokens (3 input : 1 output), log scaleArena Text (Elo)Gemini 4 CarbonGemini 4 ArgonClaude Opus 5.5Gemini 3.8 FlashMuse WonderMiMo-V2.6-ProGLM-5.3-FlashMiMo-V2.6-Flash
On the frontier:Gemini 3.8 Flash1,496 · $1.50GLM-5.3-Flash1,477 · $0.24MiMo-V2.6-Pro1,480 · $0.54MiMo-V2.6-Flash1,448 · $0.18Gemini 4 Carbon1,528 · $4
Quality against price: the data
Arena Text score against blended token price for 44 current language models.
ModelProviderBlended price per 1M tokensArena Text (Elo)
Gemini 4 CarbonGoogle$41,528
Gemini 4 ArgonGoogle$41,520
Claude Opus 5.5Anthropic$81,509
Muse WonderMeta$81,507
Claude Fable 5Anthropic$201,504
Claude Fable 5.1Anthropic$201,504
Gemini 3.8 FlashGoogle$1.501,496
Muse Spark 1.3Meta$21,495
Claude Opus 5Anthropic$101,492
Kimi K3Moonshot Kimi$61,488
GPT-5.6 SolOpenAI$81,485
Gemini 3.6 FlashGoogle$31,483
GPT-5.5OpenAI$111,482
Claude Opus 4.8Anthropic$101,482
Qwen3.8-MaxAlibaba Qwen$31,482
GPT-6.1 SolOpenAI$41,481
GLM-5.3Z.ai$2.151,481
MiMo-V2.6-ProXiaomi MiMo$0.541,480
Claude Sonnet 5.5Anthropic$41,477
GLM-5.3-FlashZ.ai$0.241,477
GPT-6 AstraOpenAI$201,476
DeepSeek-V4.1-FlashDeepSeek$0.521,476
GPT-5.4OpenAI$5.631,475
Gemini 3 Flash PreviewGoogle$1.131,473
Grok 4.20xAI$1.561,472
GPT-5.6 TerraOpenAI$4.501,467
Grok 4.5xAI$31,465
DeepSeek-V4-ProDeepSeek$1.981,465
Claude Sonnet 5Anthropic$41,462
GPT-6 SolOpenAI$41,458
Gemini 3.5 Flash-LiteGoogle$0.851,456
Grok 4.6xAI$31,455
GPT-5.6 LunaOpenAI$0.451,453
MiMo-V2.6-FlashXiaomi MiMo$0.181,448
Step 5 PreviewStepFun$1.431,448
Grok 4.3xAI$1.561,442
GPT-6 LunaOpenAI$0.201,441
Grok 4.7xAI$31,441
Mistral Large 4Mistral AI$1.051,432
Claude Haiku 5.5Anthropic$0.201,425
GPT-4.1OpenAI$3.501,415
Solar Pro 4Upstage$0.521,388
GPT-4.1 miniOpenAI$0.701,383
GPT-4.1 nanoOpenAI$0.181,322
Scores fromArena · 2026-11-03

The best score, release by release

Arena Text · measured Oct 28, 2026

The best Arena Text score has risen from 1,324 to 1,528. Gemini 4 Carbon set the current record, released Oct 28, 2026.

The best score, release by releaseArena Text score by release date for 71 language models.1,1501,2001,2501,3001,3501,4001,4501,5001,550Jul 2024Jan 2025Jul 2025Jan 2026Jul 2026Release dateArena Text (Elo)Gemini 4 Argon
Latest records:Claude Opus 5.51,509 · Sep 22, 2026Gemini 4 Argon1,520 · Sep 30, 2026Gemini 4 Carbon1,528 · Oct 28, 2026
The best score, release by release: the data
Arena Text score by release date for 71 language models.
ModelProviderReleasedArena Text (Elo)
Gemini 4 CarbonGoogleOct 28, 20261,528
Gemini 4 ArgonGoogleSep 30, 20261,520
Claude Opus 5.5AnthropicSep 22, 20261,509
Muse WonderMetaOct 12, 20261,507
Claude Fable 5AnthropicJun 9, 20261,504
Claude Fable 5.1AnthropicSep 1, 20261,504
Gemini 3.8 FlashGoogleSep 2, 20261,496
Muse Spark 1.3MetaSep 2, 20261,495
Claude Opus 5AnthropicJul 24, 20261,492
Gemini 3.7 FlashGoogleAug 13, 20261,490
Kimi K3Moonshot KimiJul 16, 20261,488
Gemini 3.1 Pro PreviewGoogleFeb 19, 20261,487
GPT-5.6 SolOpenAIJul 9, 20261,485
Gemini 3.6 FlashGoogleJul 21, 20261,483
GPT-5.5OpenAIApr 23, 20261,482
Claude Opus 4.8AnthropicMay 28, 20261,482
Qwen3.8-MaxAlibaba QwenAug 2, 20261,482
GLM-5.3Z.aiAug 18, 20261,481
GPT-6.1 SolOpenAISep 29, 20261,481
MiMo-V2.6-ProXiaomi MiMoSep 22, 20261,480
Gemini 3.5 FlashGoogleMay 19, 20261,478
GLM-5.3-FlashZ.aiAug 26, 20261,477
Claude Sonnet 5.5AnthropicSep 28, 20261,477
GPT-6 AstraOpenAISep 3, 20261,476
DeepSeek-V4.1-FlashDeepSeekSep 10, 20261,476
GPT-5.4OpenAIMar 5, 20261,475
Gemini 3 Flash PreviewGoogleDec 17, 20251,473
Grok 4.20xAIMar 10, 20261,472
GPT-5.6 TerraOpenAIJul 9, 20261,467
Grok 4.5xAIJul 16, 20261,465
DeepSeek-V4-ProDeepSeekAug 13, 20261,465
Claude Sonnet 5AnthropicJun 30, 20261,462
GPT-6 SolOpenAISep 22, 20261,458
Gemini 3.5 Flash-LiteGoogleJul 21, 20261,456
Grok 4.6xAIAug 12, 20261,455
GPT-5.6 LunaOpenAIJul 9, 20261,453
MiMo-V2.6-FlashXiaomi MiMoSep 22, 20261,448
Gemini 2.5 ProGoogleMar 25, 20251,445
Grok 4.7xAISep 21, 20261,441
GPT-6 LunaOpenAISep 22, 20261,441
Beam 1.0Reflection AIOct 16, 20261,436
Mistral Large 4Mistral AIOct 6, 20261,432
Muse GlimmerMetaAug 10, 20261,425
Claude Haiku 5.5AnthropicOct 7, 20261,425
Kimi K2 InstructMoonshot KimiJul 11, 20251,418
GPT-4.1OpenAIApr 14, 20251,415
Nemotron 3.5 Super VLNVIDIAOct 15, 20261,412
Grok 3xAIFeb 19, 20251,411
DeepSeek-R1DeepSeekJan 20, 20251,398
Claude 3.7 SonnetAnthropicFeb 24, 20251,388
Solar Pro 4UpstageAug 6, 20261,388
GPT-4.1 miniOpenAIApr 14, 20251,383
Gemini 2.0 FlashGoogleDec 11, 20241,360
DeepSeek-V3DeepSeekDec 26, 20241,358
Nemotron 3.5 LightningNVIDIAAug 11, 20261,353
GPT-4oOpenAIMay 13, 20241,346
Granite 4.2 30BIBMAug 25, 20261,340
Grok 2xAIAug 13, 20241,336
Llama 3.1 405BMetaJul 23, 20241,335
Gemini 1.5 ProGoogleFeb 15, 20241,324
Claude 3 OpusAnthropicMar 4, 20241,322
GPT-4.1 nanoOpenAIApr 14, 20251,322
Mistral Large 2Mistral AIJul 24, 20241,314
Gemini 1.5 FlashGoogleMay 14, 20241,287
Claude 3 SonnetAnthropicMar 4, 20241,281
Llama 3 70BMetaApr 18, 20241,277
DeepSeek-Coder-V2 InstructDeepSeekJun 17, 20241,265
Claude 3 HaikuAnthropicMar 13, 20241,262
Qwen2 72B InstructAlibaba QwenJun 7, 20241,262
Ministral 8BMistral AIOct 16, 20241,238
Llama 3 8BMetaApr 18, 20241,224
Scores fromArena · 2026-11-03

Each provider's best model

Arena Text · measured Oct 28, 2026

16 providers have a current model scored on Arena Text. Google leads with Gemini 4 Carbon.

  1. 1

    Google

    Gemini 4 CarbonNew
    1,528 Elo
  2. 2

    Anthropic

    Claude Opus 5.5
    1,509 Elo
  3. 3

    Meta

    Muse WonderNew
    1,507 Elo
  4. 4

    Moonshot Kimi

    Kimi K3
    1,488 Elo
  5. 5

    OpenAI

    GPT-5.6 Sol
    1,485 Elo
  6. 6

    Alibaba Qwen

    Qwen3.8-Max
    1,482 Elo
  7. 7

    Z.ai

    GLM-5.3
    1,481 Elo
  8. 8

    Xiaomi MiMo

    MiMo-V2.6-Pro
    1,480 Elo
  9. 9

    DeepSeek

    DeepSeek-V4.1-Flash
    1,476 Elo
  10. 10

    xAI

    Grok 4.20
    1,472 Elo
  11. 11

    StepFun

    Step 5 Preview
    1,448 Elo
  12. 12

    Reflection AI

    Beam 1.0New
    1,436 Elo
  13. 13

    Mistral AI

    Mistral Large 4
    1,432 Elo
  14. 14

    NVIDIA

    Nemotron 3.5 Super VLNew
    1,412 Elo
  15. 15

    Upstage

    Solar Pro 4
    1,388 Elo
  16. 16

    IBM

    Granite 4.2 30B
    1,340 Elo

Elo rating; the axis starts at 1,300, since a rating has no zero.

Scores fromArena · 2026-11-03

Fixed-price audit

Find the cheapest model that still does your job

The rankings above use public benchmarks. The AI Model Cost Audit applies the same data to your workload: your tasks, your volumes, your bill.

AI Model Cost Audit

Launch price

Launch price $100, regular price $150

One-time, fixed price

What we need: your usage export or invoices and a few sample prompts. No access to your systems.

Book the cost audit

What you get

  • A review of your current AI spend: the models you call, the task behind each call and what it costs a month.
  • A shortlist for each task: the models that score as well on the benchmarks that match it, for less.
  • A savings estimate: your monthly bill today against the recommended setup, with every assumption written down.
  • Quick wins your team can apply on its own: prompt caching, batching, shorter prompts and routing rules.
  • A 30-minute call with the founder to walk through the report.

Explore

Filter and compare any models

Choose a task, filter its models by provider, capability or status, and select up to four to compare side by side. Missing values stay visible as Not reported.

The explorer opens as you reach it. Every model also has its own page in the directory below, and the whole dataset downloads as JSON or CSV.

Model pages

Every model, one page each

Pricing, specs, independent scores, alternatives and sources for all 157 models, grouped by provider.

Google 30

OpenAI 29

xAI 15

Anthropic 13

Alibaba Qwen 8

Meta 8

ElevenLabs 7

Mistral AI 7

DeepSeek 6

Moonshot Kimi 6

Microsoft AI 4

NVIDIA 3

Inception 2

Xiaomi MiMo 2

Z.ai 2

AssemblyAI 1

Black Forest Labs 1

ByteDance Seed 1

Cartesia 1

Elorian 1

IBM 1

Meituan 1

MiniMax 1

Reflection AI 1

Runway 1

Stability AI 1

StepFun 1

Tencent 1

Thinking Machines Lab 1

Upstage 1

Head-to-head comparisons

All 652 comparisons

Methods and provenance

Evidence library

Open only the methodology, production guidance, release history, sources, or answers you need. Each stays in the page HTML while closed, and every source record is in the source catalog.

How to read benchmark data

How to read the benchmark data

Use the catalog to screen candidates, then verify them on your workload. Linked official sources support each record as a whole; they are not a field-by-field audit trail. Benchmark results retain their publisher and compatible snapshot.

Keep task families separate

Scores and operational metrics should not be compared across task families. Text reasoning, image quality, video quality, and speech error metrics answer different questions.

Match one benchmark snapshot

Compare results only when the task family, benchmark version, measurement date, and eligible model set match. Different snapshots may use different prompts, evaluators, or datasets.

Follow the score direction

Higher-is-better metrics reward a larger result; lower-is-better metrics such as error rate reward a smaller result. The direction belongs to the named benchmark, not to every task.

Match the pricing unit

A price per million tokens, image, audio minute, video second, or realtime minute is a different cost basis. Compare only the same unit and tier, then add retries, review, tooling, and infrastructure.

Check lifecycle and source status

Current models are active provider offerings; Legacy models remain for historical context, and one its provider deprecated shows the date and its replacement. Official records come from providers, while independent results come from benchmark publishers, each with a visible verification date.

Keep missing data missing

Not reported is not zero. The explorer excludes missing values from rankings and does not infer them from a related model.

Scope begins 2024-01-01. Dataset version 2026-11-06. Last verified 2026-11-06. 255 source records.

Production fit

Why a benchmark winner may not fit production

Benchmark rank and production fit measure different things. A result can support a shortlist without deciding the system design.

  • A public score may cover one narrow task while your system combines retrieval, tools, permissions, and review.
  • Production acceptance also depends on latency, reliability, safety, data handling, regional availability, and total cost.
  • Van Data Team recommendations are transparent editorial shortlists. They use published fields and results, apply no hidden weighting, and must be tested on representative work.
Release timeline

Model release timeline since 2024

Generated from exact explorer release dates. Models without a verified exact date remain in the inventory and are omitted here rather than assigned an inferred date.

  1. 2026 Q4

    • Gemini Omni 1.2 Flash2026-11-05
    • GPT-6 Cyber2026-11-05
    • Inkling 2 Small2026-11-05
    • GPT Image 2.6 Flare2026-11-04
    • GPT Image 2.6 Sunburst2026-11-04
    • Hy42026-11-04
    • LongCat-2.52026-11-02
    • GPT-Rosalind-62026-10-30
    • AssemblyAI-TTS-12026-10-28
    • Gemini 4 Carbon2026-10-28
    • GPT-6.1 Terra2026-10-26
    • Kimi K3.1 Pro2026-10-23
    • Voxtral Mini TTS 32026-10-20
    • Voxtral Small Transcribe 32026-10-20
    • Mercury 32026-10-19
    • MiniMax M3.1 Flash2026-10-19
    • Beam 1.02026-10-16
    • Glimpse2026-10-16
    • Muse Spark Open2026-10-15
    • Nemotron 3.5 Bolt 74B-A9B2026-10-15
    • Nemotron 3.5 Super VL2026-10-15
    • Grok Voice 3.02026-10-14
    • Muse Wonder2026-10-12
    • Claude Haiku 5.52026-10-07
    • Mistral Large 42026-10-06
    • Nano Banana 2.12026-10-06
    • MAI-Transcribe-2-Streaming2026-10-01
    • MAI-Voice-2.12026-10-01
    • MAI-Voice-2.1-Flash2026-10-01

  2. 2026 Q3

    • Gemini 4 Argon2026-09-30
    • GPT-6.1 Sol2026-09-29
    • Claude Sonnet 5.52026-09-28
    • Eleven v42026-09-28
    • Eleven v4 Turbo2026-09-28
    • Claude Opus 5.52026-09-22
    • Gemini 3.8 Flash TTS2026-09-22
    • Gemini 3.8 Flash-Lite TTS2026-09-22
    • GPT-6 Luna2026-09-22
    • GPT-6 Sol2026-09-22
    • MiMo-V2.6-Flash2026-09-22
    • MiMo-V2.6-Pro2026-09-22
    • Grok 4.72026-09-21
    • Qwen3.8-Omni-Flash2026-09-20
    • Grok Voice Transcribe 2.02026-09-18
    • Gemini 3.8 Live2026-09-15
    • DeepSeek-V4.1-Flash2026-09-10
    • GPT-Live 12026-09-10
    • GPT Image 2.5 Flare2026-09-08
    • GPT Image 2.5 Sunburst2026-09-08
    • Mercury 2.52026-09-08
    • GPT-6 Astra2026-09-03
    • MAI-Transcribe-22026-09-03
    • Gemini 3.8 Flash2026-09-02
    • Muse Spark 1.32026-09-02
    • Claude Fable 5.12026-09-01
    • Claude Mythos 5.12026-09-01
    • Gemini Omni Flash 1.12026-08-27
    • Sonic 3.62026-08-27
    • Gemini 3.5 Transcribe2026-08-26
    • GLM-5.3-Flash2026-08-26
    • Qwen3.8-Flash-Next2026-08-26
    • Granite 4.2 30B2026-08-25
    • GLM-5.32026-08-18
    • DeepSeek-V4-Pro2026-08-13
    • Gemini 3.7 Flash2026-08-13
    • Grok 4.62026-08-12
    • Nemotron 3.5 Lightning2026-08-11
    • GPT-5.6 Cyber2026-08-10
    • Muse Glimmer2026-08-10
    • Solar Pro 42026-08-06
    • Wan 3.0 Video2026-08-06
    • Qwen3.8-Max2026-08-02
    • GPT Transcribe2026-07-28
    • Claude Opus 52026-07-24
    • Gemini 3.5 Flash-Lite2026-07-21
    • Gemini 3.6 Flash2026-07-21
    • Grok 4.52026-07-16
    • Kimi K32026-07-16
    • GPT-5.6 Luna2026-07-09
    • GPT-5.6 Sol2026-07-09
    • GPT-5.6 Terra2026-07-09

  3. 2026 Q2

    • Claude Sonnet 52026-06-30
    • Gemini 3.1 Flash Lite Image2026-06-30
    • Gemini Omni Flash Preview2026-06-30
    • Claude Fable 52026-06-09
    • Claude Opus 4.82026-05-28
    • Gemini 3 Pro Image2026-05-28
    • Gemini 3.1 Flash Image2026-05-28
    • Gemini 3.5 Flash2026-05-19
    • Gemini 3.1 Flash-Lite2026-05-07
    • GPT-Realtime-22026-05-07
    • GPT-5.52026-04-23
    • Gemini 3.1 Flash TTS Preview2026-04-15

  4. 2026 Q1

    • Veo 3.1 Lite Preview2026-03-31
    • Gemini 3.1 Flash Live Preview2026-03-26
    • Grok 4.202026-03-10
    • GPT-5.42026-03-05
    • Gemini 3.1 Pro Preview2026-02-19
    • Scribe v22026-01-09

  5. 2025 Q4

    • Gemini 3 Flash Preview2025-12-17
    • Scribe v2 Realtime2025-11-11
    • Veo 3.1 Preview2025-10-15

  6. 2025 Q3

    • Kimi K2 Base2025-07-11
    • Kimi K2 Instruct2025-07-11

  7. 2025 Q2

    • Kimi-Dev-72B2025-06-17
    • GPT-4.12025-04-14
    • GPT-4.1 mini2025-04-14
    • GPT-4.1 nano2025-04-14
    • Kimi-VL-A3B-Instruct2025-04-09

  8. 2025 Q1

    • Gemini 2.5 Pro2025-03-25
    • GPT-4o Transcribe2025-03-20
    • Claude 3.7 Sonnet2025-02-24
    • Grok 32025-02-19
    • Qwen2.5-VL 72B Instruct2025-01-26
    • DeepSeek-R12025-01-20

  9. 2024 Q4

    • DeepSeek-V32024-12-26
    • Gemini 2.0 Flash2024-12-11
    • Ministral 8B2024-10-16

  10. 2024 Q3

    • Llama 3.2 11B Vision2024-09-25
    • Qwen2.5 32B Instruct2024-09-19
    • Pixtral 12B2024-09-17
    • Grok 22024-08-13
    • Mistral Large 22024-07-24
    • Llama 3.1 405B2024-07-23

  11. 2024 Q2

    • DeepSeek-Coder-V2 Instruct2024-06-17
    • Qwen2 72B Instruct2024-06-07
    • Qwen2 7B Instruct2024-06-07
    • Codestral 22B2024-05-29
    • Gemini 1.5 Flash2024-05-14
    • GPT-4o2024-05-13
    • DeepSeek-V2 Chat2024-05-06
    • Llama 3 70B2024-04-18
    • Llama 3 8B2024-04-18
    • Grok-1.5V2024-04-12

  12. 2024 Q1

    • Grok-1.52024-03-28
    • Claude 3 Haiku2024-03-13
    • Claude 3 Opus2024-03-04
    • Claude 3 Sonnet2024-03-04
    • Gemini 1.5 Pro2024-02-15
Source catalog · 255 records

Source catalog

Official sources come from model providers. Independent sources come from a separate benchmark publisher. The source catalog lists all 255 records, each with its publisher, verification date, version, license status, coverage and note; missing metadata is not filled or inferred.

Independent benchmark sources

Every benchmark score comes from one of these 22 records.

Arena 8
  • Arena Text-to-Image leaderboard

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People compare two anonymous images generated from the same prompt and vote for the better one. Arena fits a Bradley-Terry model to the votes and reports ratings on an Elo-like scale.

  • Arena Image Edit leaderboard

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People compare two anonymous edits of the same image from the same instruction and vote for the better one. Arena fits a Bradley-Terry model to the votes and reports ratings on an Elo-like scale.

  • Arena Text-to-Video leaderboard

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People compare two anonymous videos generated from the same prompt and vote for the better one. Arena fits a Bradley-Terry model to the votes and reports ratings on an Elo-like scale.

  • Arena Image-to-Video leaderboard

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People compare two anonymous videos animated from the same image and prompt and vote for the better one. Arena fits a Bradley-Terry model to the votes and reports ratings on an Elo-like scale.

  • Arena Text leaderboard

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People put the same prompt to two anonymous models and vote for the better answer. Arena fits a Bradley-Terry model to the votes; the rating used here is the style-controlled one, which is the default view on arena.ai.

  • Arena WebDev leaderboard (Code Arena)

    Covers benchmark · Verified 2026-11-03 · Not versioned · License CC-BY-4.0

    People give the same web development request to two anonymous models, try the apps they build and vote for the better one. Arena fits a Bradley-Terry model to the votes and reports ratings on an Elo-like scale.

The 2 records below: Ratings from blind pairwise human votes, published by Arena (formerly LMArena) in its public leaderboard dataset and pinned here to one commit. The overall category of the latest split is used; for text, the style-controlled leaderboard, which is the default view on arena.ai, and for code, the WebDev leaderboard of Code Arena. A model maps only when its Arena row names the same model; where Arena lists several quality or resolution settings of one model, the best rated row is used and recorded in sourceRow. Ratings are rounded to whole points. Credit: Arena, lmarena-ai/leaderboard-dataset on Hugging Face, CC BY 4.0.

Artificial Analysis 5
  • Artificial Analysis Speech Arena leaderboard (text to speech)

    Covers benchmark · Verified 2026-11-03 · Version Provider voice leaderboard, read 2026-11-03 · License Artificial Analysis Terms of Use

    Listeners compare two anonymous clips of the same text and vote for the more natural one; Artificial Analysis fits Elo ratings to the votes, using each provider's own voices. Cited with a link as Artificial Analysis asks; its terms reserve redistribution, so these scores are shown on the page and left out of the dataset download.

  • Artificial Analysis speech-to-text leaderboard (AA-WER v2, non-streaming)

    Covers benchmark · Verified 2026-11-03 · Version AA-WER v2, read 2026-11-03 · License Artificial Analysis Terms of Use

    Artificial Analysis transcribes the same held-out audio through each provider API and reports the weighted word error rate. Cited with a link as Artificial Analysis asks; its terms reserve redistribution, so these scores are shown on the page and left out of the dataset download.

  • Artificial Analysis speech-to-text leaderboard (AA-WER Streaming)

    Covers benchmark · Verified 2026-11-03 · Version AA-WER Streaming, read 2026-11-03 · License Artificial Analysis Terms of Use

    Artificial Analysis streams the same held-out audio through each provider realtime API and reports the word error rate of the final transcript. Cited with a link as Artificial Analysis asks; its terms reserve redistribution, so these scores are shown on the page and left out of the dataset download.

  • Artificial Analysis Video Arena leaderboard (AA-Video-T2V v2.1)

    Covers benchmark · Verified 2026-11-03 · Version AA-Video-T2V v2.1, read 2026-11-03 · License Artificial Analysis Terms of Use

    People compare two anonymous videos generated from the same prompt and vote for the better one; Artificial Analysis fits Elo ratings to the votes. Cited with a link as Artificial Analysis asks; its terms reserve redistribution, so these scores are shown on the page and left out of the dataset download.

  • Artificial Analysis Video Arena leaderboard (AA-Video-I2V v2.1)

    Covers benchmark · Verified 2026-11-03 · Version AA-Video-I2V v2.1, read 2026-11-03 · License Artificial Analysis Terms of Use

    People compare two anonymous videos animated from the same image and prompt and vote for the better one; Artificial Analysis fits Elo ratings to the votes. Cited with a link as Artificial Analysis asks; its terms reserve redistribution, so these scores are shown on the page and left out of the dataset download.

Epoch AI 4
  • GPQA Diamond benchmark (Epoch AI)

    Covers benchmark · Verified 2026-10-12 · Not versioned · License CC-BY-4.0

    198 four-option graduate-level questions in biology, physics and chemistry, written by domain experts; random guessing scores 25%. Epoch runs the evaluation itself.

  • FrontierMath benchmark (Epoch AI)

    Covers benchmark · Verified 2026-10-12 · Not versioned · License CC-BY-4.0

    Original research-level mathematics problems written and vetted by expert mathematicians. Tiers 1-3 v2 (released 2026-06-12) is the 295-problem base set after a revision that corrected errors in 42% of problems. Scores come from Epoch-run private sets. OpenAI funded the benchmark and has access to part of it.

  • SimpleQA Verified benchmark (Epoch AI)

    Covers benchmark · Verified 2026-10-12 · Not versioned · License CC-BY-4.0

    A 1,000-question factoid benchmark from Google, building on SimpleQA from OpenAI. Epoch runs it itself and reports the proportion answered correctly, not the F1-based score Google reports.

  • Epoch AI Benchmarking Hub data (benchmark_data.zip)

    Covers benchmark · Verified 2026-11-03 · Version benchmark_data.zip retrieved 2026-11-03, sha256 prefix 333c5aa3e36a4ea1 · License CC-BY-4.0

    Results Epoch AI ran itself (the GPQA Diamond, FrontierMath Tiers 1-3 v2 and SimpleQA Verified files), not the external files it collects from other publishers. A model maps only when its Epoch row is unambiguous: exact name, or a dated version pinned by the record release date. Where Epoch lists several reasoning-effort settings, the best published configuration is used and its row is recorded in sourceRow. Scores are converted from 0-1 to 0-100 and rounded to one decimal. Epoch ran the newest rows up to 2026-11-02; every result is dated to this capture. Credit: Epoch AI, Capabilities and benchmarking, epoch.ai, CC BY 4.0.

LiveBench 4
  • LiveBench category-weighted averaging implementation

    Covers benchmark · Verified 2026-08-07 · Version LiveBench-2026-06-25@19e766a5de4de07d672ed5bf9f0a69ceed1d39bf · License Apache-2.0

    Pinned implementation averages task scores within each of the seven categories, then averages the category scores; the paired categories_2026_06_25.json file defines category membership.

  • LiveBench 2026-06-25 task results

    Covers benchmark · Verified 2026-08-07 · Version LiveBench-2026-06-25@19e766a5de4de07d672ed5bf9f0a69ceed1d39bf · License Apache-2.0

    Snapshot and measurement date: 2026-06-25. Overall scores are imported only for unambiguous model mappings and are recomputed with the pinned category-weighted averaging implementation.

  • LiveBench datasheet and methodology

    Covers benchmark · Verified 2026-08-06 · Version LiveBench-2024-11-25 · License Apache-2.0

    Primary benchmark documentation confirms objective scoring, public distribution, and Apache 2.0 reuse terms.

  • LiveBench 2024-11-25 task results

    Covers benchmark · Verified 2026-08-06 · Version LiveBench-2024-11-25@347e3d0b6a4916cd66f8bda31ce7ecdd7436ecc3 · License Apache-2.0

    Snapshot and measurement date: 2024-11-25. Exact published snapshot; only direct Web of Lies V2 percentages with unambiguous model mappings are imported.

OpenRouter 1
  • OpenRouter models API catalogue

    Covers identity, pricing · Verified 2026-09-09 · Not versioned · License not stated

    Third-party API catalogue used only to cross-check published token prices, context limits, and listing dates against first-party documentation.

Official provider sources

233 records from 35 publishers, each listed with every field in the source catalog:

All 255 source records
FAQ

AI model benchmark FAQ

What does an AI model benchmark score mean?

A benchmark score reports performance on one defined evaluation. Compare scores only when the task family, benchmark version, measurement date, scoring method, and eligible model set match; the result is a screening signal, not a prediction for every workload.

How fresh are the model and source records?

Each source shows a verification date, and the page shows the dataset version and last verified date. A verification date confirms when Van Data Team checked the linked record; it does not promise that a provider page has stayed unchanged since then.

Why are some model values marked Not reported?

Not reported means the catalog does not have a compatible, directly sourced value for that exact model record. Missing values remain empty rather than being estimated, copied from a related model, or treated as zero.

How should I compare AI model pricing?

Compare only the same pricing unit and tier, such as a million tokens, image, audio minute, video second, character, or realtime minute. Prices are provider list prices checked on the source verification date, not live quotes; retries, review, tooling, and infrastructure also affect production cost.

Does open-weight access mean a model is free to run?

No. Open weights means the model parameters are available for deployment under the provider license. This can provide infrastructure control, but compute, serving, monitoring, security, and engineering still carry costs.

How should I choose models for a production evaluation?

Start with the use case, select a small evidence-backed shortlist, and test it on representative work. Set acceptance thresholds for quality, latency, cost, reliability, safety, and human review before comparing results.

Where do the benchmark scores come from, and how often are they updated?

Only from independent publishers: Epoch AI (GPQA Diamond, FrontierMath and SimpleQA Verified), LiveBench, the Arena leaderboards for text, images and video, and Artificial Analysis for speech and video. Vendor launch numbers are never mixed in. Where a model was tested at several reasoning-effort settings, the best published setting is shown with its exact leaderboard row. The dataset is updated when a review verifies a material model, price or benchmark change, and the last verified date records it.

How can Van Data Team help with model evaluation?

Van Data Team can turn a shortlist into a production evaluation with representative cases, measurable acceptance rules, model routing, workflow controls, observability, and cost tracking tailored to your system.

Production evaluation

Turn a model shortlist into a production decision

We design representative evaluations and the workflow controls needed to operate the selected model with clear quality, cost, and review boundaries.

  • Evaluation cases and acceptance thresholds tied to real work
  • Model routing, tool permissions, observability, and human review design
Review the methodologyBook a consultation