Skip to content

AI Model Rankings & Benchmarks

245 text and coding models, ranked. Context window for 235, input price for 218 and an LMArena Elo for 224 — every figure sourced from the vendor or LMArena, with gaps shown as —. How these rankings are built · Open-weight models only

333

Models tracked, all types

58

Providers

10

Published benchmarks

Nov 1, 2026

Data Refreshed

Top LMArena Elo
1528
Gemini 4 Carbon (High)
Google DeepMind · arena.ai, verified
See it in the full leaderboard
Cheapest Published Price
$0/1M in
Gemma 4 31B
Google · vendor-published
Fastest Measured
842 tokens/s
gpt-oss-120b on Cerebras
OpenAI · fastest provider on OpenRouter, 30-min median · 141 of 245 models measured

Full Leaderboard

Every column is sourced from the vendor or LMArena - sort any column, or search the 242 rows below (one per model, or per effort where LMArena ranks several). LMArena Elo shows each row's own retrieval date; the board moves weekly. ± is LMArena's 95% confidence interval. Rank is LMArena's published rank, which counts entries not listed here.

242 rows, one per model or per LMArena-ranked effort. LMArena rank, LMArena Elo, input price per million tokens and context window. Sortable by any column.
1Gemini 4 Carbon (High)1528±10high effort$21M tokens
2Gemini 4 Argon (High)1520±7high effort$21,048,576 tokens
3Claude Opus 5.5 (High)1509±7high effort$41M tokens
4Muse Wonder (Max)1507±7max effort$41M tokens
5Claude Fable 5.1 (Max)1504±6max effort$101M tokens
6Claude Fable 5 (High)1504±4high effort$101M tokens
7Claude Opus 4.6 (High)1503±3high effort$51M tokens
8Claude Opus 4.7 (High)1501±4high effort$51M tokens
9Claude Opus 4.6 (Default)1496±3default effort$51M tokens
10Gemini 3.8 Flash (High)1496±4high effort$0.751M tokens
1-10 of 242
Page 1 of 25

Agent Arena

Net improvement on agentic tasks, against an average model

LMArena Agent Arena, CC-BY · rank is LMArena’s · higher is better · 0% is an average model · top 10 of the 60 rated models carried here

60 AI models by LMArena Agent Arena net improvement, in percentage points against an average model. Positive is above average, negative below.
RankModelNet
1Claude Opus 5.5 (max)+20.9%±2.15
2Gemini 4 Carbon (high)+16%±1
3Claude Sonnet 5.5 (max)+13.8%±2.65
4Claude Fable 5.1 (max)+13.4%±1.85
5GPT-6 Astra (max)+13%±2.05
6GPT-6.1 Sol (max)+12.5%±1.05
7Gemini 4 Argon (high)+10.7%±0.8
10Claude Fable 5 (high)+8%±1.1
11GPT-6 Sol (max)+8%±1.65
11Claude Opus 4.8 (high)+7.1%±1.55

LMArena category ranks

Rank within each category, out of every model LMArena rates

LMArena, CC-BY · lower is better · sorted by Overall

224 AI models ranked by LMArena in 8 categories. First, second and third place are shaded; the rank is printed in every cell. An em dash means LMArena does not rank that model in that category.
Model
Gemini 4 Carbon11113311
Gemini 4 Argon24251222
Claude Opus 5.535339133
Muse Wonder42489329811
Claude Fable 5.15109152878
Claude Fable 562546454
Claude Opus 4.69867151496
Gemini 3.8 Flash10161122410149
Muse Spark 1.31119141711252121
Claude Opus 4.712910829121210
Muse Spark 1.21355161028582930
Claude Opus 514613165161012
Muse Spark 1.11639191925523548
Gemini 3.7 Flash172120381861716
Muse Spark1860232560264654
1–15 of 224
Loading benchmark charts…

Benchmark leaders

The top 3 text models on LMArena's Agent Arena and each of the 10 published benchmarks charted above, with the score each result gives. Higher is better on every one. Each row notes the release date of the newest model that benchmark has rated, which shows how current its results are.

  1. Agent Arena

    LMArena · 60 models rated · newest released Oct 2026

    1. 1Claude Opus 5.5 (max)+20.9% · rank 1
    2. 2Gemini 4 Carbon (high)+16% · rank 2
    3. 3Claude Sonnet 5.5 (max)+13.8% · rank 3

  2. GPQA Diamond

    Epoch AI · 67 models rated · newest released Oct 2026

    1. 1Gemini 4 Argon (high)96%
    2. 2Claude Sonnet 5.5 (max)96%
    3. 3GPT-6 Astra96%

  3. SimpleBench

    SimpleBench · 25 models rated · newest released Aug 2026

    1. 1Claude Fable 5 (max)82%
    2. 2Muse Spark 1.275%
    3. 3Qwen3.7 Max70%

  4. LiveBench

    LiveBench · 7 models rated · newest released Dec 2024

    1. 1DeepSeek R172%
    2. 2QwQ 32B72%
    3. 3Gemma 3 27B IT50%

  5. SWE-bench Verified

    Epoch AI · 14 models rated · newest released Aug 2026

    1. 1Claude Opus 4.7 (max)83%
    2. 2Claude Opus 4.679%
    3. 3Gemini 3.5 Flash79%

  6. Aider Polyglot

    Aider · 8 models rated · newest released Aug 2025

    1. 1Grok 4 070980%
    2. 2Qwen3 235B A22b Instruct 250760%
    3. 3DeepSeek R157%

  7. FrontierMath

    Epoch AI · 42 models rated · newest released Oct 2026

    1. 1GPT-6.1 Sol (max)94%
    2. 2GPT-6 Astra (max)94%
    3. 3Claude Opus 5.5 (max)91%

  8. MATH Level 5

    Epoch AI · 14 models rated · newest released Dec 2024

    1. 1DeepSeek R193%
    2. 2Mistral Medium 250582%
    3. 3Gemma 3 27B IT74%

  9. Mock AIME 2024-25

    Epoch AI · 67 models rated · newest released Sep 2026

    1. 1Gemini 4 Argon (high)100%
    2. 2Claude Opus 5.5 (max)100%
    3. 3Claude Fable 5.1100%

  10. SimpleQA Verified

    Epoch AI · 45 models rated · newest released Oct 2026

    1. 1Gemini 4 Argon (high)76%
    2. 2GPT-6 Astra76%
    3. 3GPT-6.1 Sol (max)74%

  11. Creative Writing

    Lech Mazur · 12 models rated · newest released Aug 2025

    1. 1Gemini 2.5 Pro84%
    2. 2DeepSeek R183%
    3. 3Qwen3 235B A22b Thinking 250782%

AI model rankings by task

The top 5 text models in each of LMArena's category leaderboards, by category Elo. A category Elo counts only the votes on prompts of that kind, so the order can differ from the overall ranking.

Coding

LMArena category Elo · 224 models rated

  1. 1Gemini 4 Carbon (high)1573
  2. 3Claude Opus 5.5 (high)1551
  3. 5Gemini 4 Argon (high)1550
  4. 4Claude Fable 5 (high)1550
  5. 7Claude Opus 4.61546

Creative Writing

LMArena category Elo · 224 models rated

  1. 1Claude Opus 5.5 (high)1520
  2. 2Gemini 4 Argon (high)1517
  3. 3Gemini 4 Carbon (high)1515
  4. 4Claude Fable 5 (high)1501
  5. 6Gemini 3.7 Flash (high)1490

Instruction Following

LMArena category Elo · 224 models rated

  1. 1Gemini 4 Carbon (high)1525
  2. 2Gemini 4 Argon (high)1525
  3. 3Claude Opus 5.5 (high)1522
  4. 5Claude Fable 5 (high)1508
  5. 7Claude Fable 5.1 (max)1500

Hard Prompts

LMArena category Elo · 224 models rated

  1. 1Gemini 4 Carbon (high)1548
  2. 2Gemini 4 Argon (high)1544
  3. 3Claude Opus 5.5 (high)1535
  4. 5Claude Fable 5 (high)1529
  5. 6Claude Opus 4.61526

Multi-Turn

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Carbon (high)1559
  2. ·Gemini 4 Argon (high)1539
  3. ·Muse Wonder (max)1516
  4. ·Claude Fable 5 (high)1516
  5. ·Claude Opus 4.71514

Longer Query

LMArena category Elo · 223 models rated

  1. 1Gemini 4 Carbon (high)1542
  2. 2Gemini 4 Argon (high)1539
  3. 3Claude Opus 5.5 (high)1525
  4. 4Claude Fable 5 (high)1522
  5. 6Claude Opus 4.61516

Expert

LMArena category Elo · 218 models rated

  1. 1Gemini 4 Carbon (high)1581
  2. 2Claude Fable 5 (high)1550
  3. 4Gemini 4 Argon (high)1541
  4. 5Claude Opus 5.5 (high)1538
  5. 6Claude Opus 5 (high)1537

Software & IT (occupational)

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Carbon (high)1552
  2. ·Gemini 4 Argon (high)1550
  3. ·Claude Opus 5.5 (high)1538
  4. ·Claude Fable 5 (high)1538
  5. ·Muse Wonder (max)1535

Writing & Literature (occupational)

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Carbon (high)1529
  2. ·Claude Opus 5.5 (high)1519
  3. ·Gemini 4 Argon (high)1517
  4. ·Claude Fable 5 (high)1508
  5. ·Claude Fable 5.1 (max)1497

Science (occupational)

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Argon (high)1536
  2. ·Gemini 4 Carbon (high)1530
  3. ·Claude Fable 5 (high)1527
  4. ·Kimi K3 (max)1524
  5. ·Claude Fable 5.1 (max)1523

Entertainment & Media (occupational)

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Argon (high)1512
  2. ·Gemini 4 Carbon (high)1509
  3. ·Claude Opus 5.5 (high)1501
  4. ·Claude Fable 5 (high)1493
  5. ·Muse Wonder (max)1487

Business & Management (occupational)

LMArena category Elo · 224 models rated

  1. ·Gemini 4 Carbon (high)1515
  2. ·Gemini 4 Argon (high)1512
  3. ·Muse Wonder (max)1502
  4. ·Claude Fable 5 (high)1501
  5. ·Muse Spark 1.2 (xhigh)1501

Mathematical (occupational)

LMArena category Elo · 220 models rated

  1. ·GPT-6.1 Sol (max)1539
  2. ·Gemini 4 Argon (high)1538
  3. ·Claude Fable 5.1 (max)1536
  4. ·Claude Opus 5 (high)1532
  5. ·Claude Fable 5 (high)1521

Medicine & Healthcare (occupational)

LMArena category Elo · 220 models rated

  1. ·Gemini 4 Argon (high)1539
  2. ·Muse Wonder (max)1530
  3. ·Claude Fable 5.1 (max)1528
  4. ·Gemini 4 Carbon (high)1518
  5. ·Claude Opus 5.5 (high)1512

What these rankings compare

This page compares 245 current AI models on figures that can be checked: API pricing and context window from each vendor, LMArena human-preference Elo, and ten published benchmarks. The Full Leaderboard above lists every model individually, ranked by Elo. Nothing on this page is our own scoring.

How these rankings are built

The ranking order is LMArena Elo: a rating built from blind votes, where people compare two anonymous models' answers and pick one. LMArena publishes each rating with a 95% confidence interval, shown as ± beside the Elo. Where a model was rated at a named reasoning effort, such as “high” or “max”, that effort is shown beside its figure.

The percentages in the charts above are published benchmark results, each from a named source: GPQA Diamond, SWE-bench Verified, FrontierMath, MATH Level 5, Mock AIME 2024-25 and SimpleQA Verified are run by Epoch AI and released under CC-BY; SimpleBench, LiveBench, Aider Polyglot and Creative Writing come from their own public leaderboards. Each benchmark name links to the page its figures were read from.

They are comparable down a column - between models on the same test - and not across columns, because each benchmark measures something different and scores on its own scale. A model with no bar has not been measured on that benchmark rather than scoring zero on it.

For beginners: a benchmark result records how a model scored on one specific test, under the conditions its publisher set. Each benchmark's source page describes what it measures.

Lowest, Largest and Highest-Ranked

Each figure comes from a vendor pricing page, a model card or LMArena.

Lowest cost per token

Gemma 4 31B

$0 / $0 per million tokens - the lowest first-party vendor API rate on this page

Largest single input

Llama 4 Scout

10M tokens context window, the largest of any model on this page

Highest LMArena rank

Gemini 4 Carbon

Elo 1528, rank 1 - the highest of the 224 models on this page with a verified leaderboard entry

Published open weights

GLM 5, HY3, DeepSeek V4-Pro, Gemma 4 31B, Kimi K2.5, DeepSeek V4 Flash High Preview, Gemma 4 26B A4b, GLM 4.7, GLM-5.3, Inkling and 106 more

Highest LMArena-rated of the 116 models on this page with published open weights

Single-GPU deployment

Nemotron 3.5 Lightning

30B parameters with 3B active, documented by NVIDIA as running on a single H100 or A100 80GB

Read Individual Model Guides

Gemini 4 Carbon Complete Guide Gemini 4 Argon Complete Guide Claude Opus 5.5 Complete Guide Muse Wonder Complete Guide Claude Fable 5 Complete Guide Claude Fable 5.1 Complete Guide
View all 245 models

Frequently Asked Questions

Which model has the lowest API price?

Gemma 4 31B, at $0 input / $0 output per million tokens - the lowest first-party vendor API price among the 245 models on this page. Note DeepSeek's rates also vary by time of day (peak 01:00-04:00 and 06:00-10:00 UTC, half price outside those windows).

What does the ± beside an Elo rating mean?

LMArena publishes each Elo rating with a 95% confidence interval, shown here as ± points. It reflects how many blind votes a model has received: a newer model with fewer votes has a wider range. Where two models’ ranges overlap, LMArena’s votes so far do not separate them.

Which model has the largest context window?

Llama 4 Scout, at 10M tokens - the largest of any model covered here. Models whose vendor publishes no context window figure are left out of this comparison.

Which model ranks highest on LMArena?

Gemini 4 Carbon, at 1528 (rank 1) - the highest of any model on this page with a leaderboard entry. 224 of the 245 models here have a verified LMArena entry; the rest have no entry we can confirm (including newly-released models), so we leave those blank rather than estimating.

Which models can I actually self-host?

116 models here publish open weights, topped by GLM 5 (MIT), HY3 (Apache 2.0), DeepSeek V4-Pro (MIT), Gemma 4 31B (Apache 2.0), Kimi K2.5 (Modified MIT), DeepSeek V4 Flash High Preview (MIT), Gemma 4 26B A4b (Apache 2.0), GLM 4.7 (MIT), GLM-5.3 (custom: MIT-style terms plus a security-review clause for MaaS providers above $10B revenue), Inkling (Apache 2.0) and 106 more. Everything else here is closed or API-only.

Which model is preview vs. stable?

Gemini 3.1 Pro (preview), Grok 4.20 Beta 0309 Reasoning (beta checkpoint; served on the xai api as ga grok-4.20-0309-reasoning.), Grok 4.20 Beta1 (beta checkpoint; superseded on the xai api by ga grok-4.20-0309-reasoning and grok-4.20-0309-non-reasoning.), Grok 4.20 Multi Agent Beta 0309 (beta checkpoint; served on the xai api as grok-4.20-multi-agent-0309.), Hunyuan HY3 Preview (preview; superseded by tencent's hy3, open-sourced under apache 2.0 after the late-april preview.), Hy4 preview (preview), Kimi K2.8 Preview (preview), Laguna M.1 (preview), Laguna XS.2 (preview), LongCat-2.0 (preview) and 2 more are not documented as stable releases. The other 233 models covered here are documented as regular releases by their vendors.

Where do the benchmark percentages come from?

Each chart is one published benchmark, as a percentage: GPQA Diamond, SWE-bench Verified, FrontierMath, MATH Level 5, Mock AIME 2024-25 and SimpleQA Verified from Epoch AI; SimpleBench, LiveBench, Aider Polyglot and Creative Writing from their own public leaderboards. They are comparable between models on the same benchmark, not across benchmarks - 62 on GPQA Diamond and 62 on Aider Polyglot are two different tests. Nothing in these charts is our own scoring.