AI Leaderboard
The best AI for each job right now, picked by people in blind head-to-head votes. Lab benchmarks have their own tab. Updated October 18, 2026.
| Job | People's #1 | Runners-up | Most voted |
|---|---|---|---|
| Chat and everyday questions8,834,211 votes | Muse WonderMeta · tied with 5 others | Claude Opus 5.5AnthropicClaude Fable 5Anthropic | Claude Opus 4.681,394 votes |
| Coding and building apps891,922 votes | Claude Opus 5.5Anthropic | GPT 6 AstraOpenAIClaude Sonnet 5.5Anthropic | Kimi K319,220 votes |
| Writing1,414,468 votes | Claude Opus 5.5Anthropic · tied with 2 others | Claude Fable 5AnthropicClaude Opus 4.6Anthropic | Gemini 3.1 Pro Preview23,619 votes |
| Web search and research1,437,508 votes | GPT-5.6 SolOpenAI · tied with 2 others | Qwen3.8-MaxAlibabaClaude Opus 4.6Anthropic | Claude Sonnet 4.6173,986 votes |
| Reading images and documents1,462,389 votes | Claude Fable 5Anthropic · tied with 7 others | Qwen3.8-MaxAlibabaClaude Opus 4.6Anthropic | Claude Opus 4.723,165 votes |
| Creating images6,604,125 votes | GPT-Image-2OpenAI | MAI-Image-2.6Microsoft AIReve 2.1Reve | Nano Banana Pro (Gemini 3 Pro Image 2k)175,734 votes |
| Editing images30,985,932 votes | GPT-Image-2OpenAI | MAI-Image-2.6Microsoft AIGrok Imagine Image 2.0SpaceXSI (formerly xAI) | GPT Image 1.5 High Fidelity654,249 votes |
| Creating video718,577 votes | Gemini Omni 1.1 FlashGoogle · tied with Gemini Omni Flash | Gemini Omni FlashGoogleDreamina Seedance 2.0ByteDance | Dreamina Seedance 2.056,318 votes |
| Turning images into video2,081,953 votes | MiniMax H3MiniMax · tied with 2 others | Gemini Omni 1.1 FlashGoogleWan 3.0Alibaba | Grok Imagine Video554,050 votes |
Full rankings
Chat and everyday questions. Blind head-to-head votes from people comparing two anonymous answers, across all kinds of questions.
| # | Model | People's score | Votes | API price, 1M tokens |
|---|---|---|---|---|
| 1= | Muse WonderMeta | 1,509 | 3,247 | $4 / $20 |
| 1= | Claude Opus 5.5Anthropic | 1,508 | 8,300 | $4 / $20 |
| 1= | Claude Fable 5Anthropic | 1,504 | 42,004 | - |
| 1= | Claude Opus 4.6Anthropic | 1,503 | 81,394 | - |
| 1= | Claude Fable 5.1Anthropic | 1,502 | 14,590 | $10 / $50 |
| 1= | Claude Opus 4.7Anthropic | 1,501 | 65,116 | - |
| 2= | Muse Spark 1.3Meta | 1,494 | 17,561 | $1.25 / $4.25 |
| 2= | Muse Spark 1.2Meta | 1,493 | 8,248 | - |
| 7= | Muse Spark 1.1Meta | 1,490 | 42,810 | - |
| 7= | Claude Opus 5Anthropic | 1,490 | 32,495 | - |
| 7= | Muse SparkMeta | 1,489 | 14,150 | - |
| 7= | Gemini 3.7 FlashGoogle | 1,488 | 27,232 | - |
Source: LMArena Text Arena (overall, style control), data from October 15, 2026, 8,834,211 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): Gemini 4 Argon, Gemini 3.8 Flash.
Coding and building apps. Blind votes on building working web apps and sites from a prompt.
| # | Model | People's score | Votes | API price, 1M tokens |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 1,807 | 2,775 | $4 / $20 |
| 2= | GPT 6 AstraOpenAI | 1,777 | 6,014 | $10 / $50 |
| 2= | Claude Sonnet 5.5Anthropic | 1,768 | 3,182 | $2 / $10 |
| 2= | GPT 6.1 SolOpenAI | 1,756 | 2,863 | $2 / $10 |
| 4 | Claude Fable 5.1Anthropic | 1,741 | 7,503 | $10 / $50 |
| 6= | Claude Opus 5Anthropic | 1,689 | 18,310 | - |
| 6= | GPT 6 SolOpenAI | 1,683 | 4,364 | - |
| 6= | Gemini 4 ArgonGoogle | 1,678 | 1,775 | $2 / $10 |
| 7 | Qwen3.8 MaxAlibaba | 1,674 | 13,812 | $2 / $6 |
| 9 | Muse Spark 1.3Meta | 1,658 | 9,068 | $1.25 / $4.25 |
| 10 | Kimi K3Moonshot AIOpen weights | 1,655 | 19,220 | $3 / $15 |
| 12 | Grok 4.7SpaceXSI (formerly xAI) | 1,639 | 5,021 | $2 / $6 |
Source: LMArena WebDev Arena, data from October 16, 2026, 891,922 votes in this arena.
Writing. Blind votes on creative writing: stories, poems, scripts and copy.
| # | Model | People's score | Votes | API price, 1M tokens |
|---|---|---|---|---|
| 1= | Claude Opus 5.5Anthropic | 1,519 | 1,945 | $4 / $20 |
| 1= | Claude Fable 5Anthropic | 1,501 | 9,020 | - |
| 1= | Claude Opus 4.6Anthropic | 1,500 | 14,950 | - |
| 2= | Claude Opus 4.7Anthropic | 1,489 | 11,806 | - |
| 2= | Muse WonderMeta | 1,485 | 710 | $4 / $20 |
| 2= | Claude Fable 5.1Anthropic | 1,485 | 3,284 | $10 / $50 |
| 4= | Gemini 3 ProGoogle | 1,485 | 6,543 | - |
| 4= | Gemini 3.1 Pro PreviewGoogle | 1,481 | 23,619 | - |
| 5= | Claude Opus 5Anthropic | 1,471 | 14,918 | - |
| 5= | Gemini 3.6 Flash HighGoogle | 1,471 | 7,583 | - |
| 5= | Claude Opus 4 5 20251101 High 32kAnthropic | 1,469 | 5,688 | - |
| 7 | Claude Opus 4 8 HighAnthropic | 1,469 | 12,797 | - |
Source: LMArena Text Arena (creative writing, style control), data from October 15, 2026, 1,414,468 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): Gemini 4 Argon, Gemini 3.7 Flash, Gemini 3.8 Flash.
Web search and research. Blind votes on answers that search the web, with sources.
| # | Model | People's score | Votes | API price, 1M tokens |
|---|---|---|---|---|
| 1= | GPT-5.6 SolOpenAI | 1,261 | 60,437 | - |
| 1= | Qwen3.8-MaxAlibaba | 1,261 | 1,870 | $2 / $6 |
| 1= | Claude Opus 4.6Anthropic | 1,251 | 173,191 | - |
| 2= | GPT-5.5OpenAI | 1,247 | 125,957 | - |
| 2= | Claude Fable 5.1Anthropic | 1,246 | 12,782 | $10 / $50 |
| 2= | Claude Opus 5Anthropic | 1,242 | 13,285 | - |
| 5= | GPT-6 AstraOpenAI | 1,234 | 18,483 | $10 / $50 |
| 5= | GPT-6.1 SolOpenAI | 1,233 | 5,555 | $2 / $10 |
| 6= | Claude Opus 4.7Anthropic | 1,230 | 127,506 | - |
| 6= | Claude Fable 5Anthropic | 1,229 | 73,596 | - |
| 6= | ERNIE 5.1Baidu | 1,227 | 4,024 | - |
| 9 | Claude Sonnet 4.6Anthropic | 1,221 | 173,986 | - |
Source: LMArena Search Arena, data from October 13, 2026, 1,437,508 votes in this arena.
Reading images and documents. Blind votes on questions about images, screenshots and documents.
| # | Model | People's score | Votes | API price, 1M tokens |
|---|---|---|---|---|
| 1= | Claude Fable 5Anthropic | 1,308 | 14,175 | - |
| 1= | Qwen3.8-MaxAlibaba | 1,301 | 10,017 | $2 / $6 |
| 1= | Claude Opus 4.6Anthropic | 1,299 | 22,304 | - |
| 1= | Claude Opus 4.7Anthropic | 1,298 | 23,165 | - |
| 1= | GPT-6.1 SolOpenAI | 1,294 | 1,813 | $2 / $10 |
| 1= | Muse SparkMeta | 1,294 | 5,872 | - |
| 1= | Muse Spark 1.2Meta | 1,293 | 1,954 | - |
| 1= | Claude Opus 5.5Anthropic | 1,293 | 2,508 | $4 / $20 |
| 2= | Muse Spark 1.3Meta | 1,290 | 4,550 | $1.25 / $4.25 |
| 2= | Gemini 3 ProGoogle | 1,289 | 13,639 | - |
| 2= | Gemini 3.8 Flash HighGoogle | 1,289 | 3,792 | - |
| 2= | Claude Opus 5 HighAnthropic | 1,287 | 16,665 | - |
Source: LMArena Vision Arena (style control), data from October 9, 2026, 1,462,389 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): Gemini 3.7 Flash.
Creating images. Blind votes on images generated from the same prompt.
| # | Model | People's score | Votes | Per 1,000 images |
|---|---|---|---|---|
| 1 | GPT-Image-2OpenAI | 1,383 | 93,334 | - |
| 2 | MAI-Image-2.6Microsoft AI | 1,332 | 25,091 | $39 |
| 3= | Reve 2.1Reve | 1,302 | 8,249 | - |
| 3= | Grok Imagine Image 2.0SpaceXSI (formerly xAI) | 1,294 | 14,483 | $60 |
| 5= | Muse ImageMeta | 1,274 | 42,262 | $10 |
| 5= | Reve 2.0Reve | 1,269 | 15,773 | - |
| 6 | Nano Banana 2 (Gemini 3.1 Flash Image)Google | 1,262 | 60,515 | - |
| 7= | Seedream 5.0 ProByteDance | 1,256 | 94,181 | $90 |
| 7= | Qwen Image 3.0 ProAlibaba | 1,255 | 11,440 | - |
| 7= | MAI Image 2.5Microsoft AI | 1,254 | 61,265 | - |
| 8 | Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)Google | 1,250 | 17,804 | - |
| 9 | Nano Banana Pro (Gemini 3 Pro Image 2k)Google | 1,248 | 175,734 | - |
Source: LMArena Text-to-Image Arena, data from October 7, 2026, 6,604,125 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, Grok Imagine Image 2.0, Gemini Nano Banana 2.1.
Editing images. Blind votes on edits made to the same photo from the same instruction.
| # | Model | People's score | Votes | Per 1,000 images |
|---|---|---|---|---|
| 1 | GPT-Image-2OpenAI | 1,462 | 305,825 | - |
| 2= | MAI-Image-2.6Microsoft AI | 1,428 | 51,384 | $39 |
| 2= | Grok Imagine Image 2.0SpaceXSI (formerly xAI) | 1,425 | 50,427 | $60 |
| 4= | Muse ImageMeta | 1,403 | 142,644 | $10 |
| 4= | MAI-Image-2.5Microsoft AI | 1,401 | 190,081 | - |
| 6= | Seedream 5.0 ProByteDance | 1,394 | 282,865 | $90 |
| 6= | Grok Imagine ImageSpaceXSI (formerly xAI) | 1,391 | 38,195 | - |
| 6= | Nano Banana Pro (Gemini 3 Pro Image 2k)Google | 1,390 | 651,674 | - |
| 6= | Chatgpt Image Latest High FidelityOpenAI | 1,389 | 629,464 | - |
| 6= | Nano Banana 2 (Gemini 3.1 Flash Image)Google | 1,387 | 229,782 | - |
| 11= | Reve 2.1Reve | 1,374 | 20,792 | - |
| 11= | GPT Image 1.5 High FidelityOpenAI | 1,370 | 654,249 | - |
Source: LMArena Image Edit Arena, data from October 6, 2026, 30,985,932 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, Grok Imagine Image 2.0, Gemini Nano Banana 2.1.
Creating video. Blind votes on videos generated from the same prompt.
| # | Model | People's score | Votes | Per minute |
|---|---|---|---|---|
| 1= | Gemini Omni 1.1 FlashGoogle | 1,516 | 1,784 | - |
| 1= | Gemini Omni FlashGoogle | 1,513 | 26,576 | $9.12 |
| 3= | Dreamina Seedance 2.0ByteDance | 1,479 | 56,318 | - |
| 3= | Wan 3.0Alibaba | 1,476 | 2,598 | $12 |
| 3= | Dreamina Seedance 2.5ByteDance | 1,474 | 8,001 | - |
| 4= | MiniMax H3MiniMaxOpen weights | 1,460 | 11,417 | $4.80 |
| 4= | Muse VideoMeta | 1,456 | 2,188 | - |
| 8 | HappyHorse 1.0Aorizon | 1,427 | 22,264 | - |
| 9= | Sora 2 ProOpenAI | 1,368 | 54,167 | - |
| 9= | Veo 3.1Google | 1,364 | 13,741 | $24 |
| 9= | Veo 3.1 Audio 1080pGoogle | 1,362 | 25,076 | - |
| 9= | Veo 3.1 Fast AudioGoogle | 1,362 | 39,484 | - |
Source: LMArena Text-to-Video Arena, data from September 22, 2026, 718,577 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): FLUX 3 Video, Grok Imagine Video 1.5.
Turning images into video. Blind votes on videos made from the same starting image.
| # | Model | People's score | Votes | Per minute |
|---|---|---|---|---|
| 1= | MiniMax H3MiniMaxOpen weights | 1,495 | 57,112 | $4.80 |
| 1= | Gemini Omni 1.1 FlashGoogle | 1,489 | 3,734 | - |
| 1= | Wan 3.0Alibaba | 1,480 | 3,442 | $12 |
| 2= | Dreamina Seedance 2.5ByteDance | 1,477 | 15,729 | - |
| 2= | Dreamina Seedance 2.0ByteDance | 1,475 | 121,427 | - |
| 3 | Gemini Omni FlashGoogle | 1,465 | 87,155 | $9.12 |
| 6 | Grok Imagine Video 1.5SpaceXSI (formerly xAI) | 1,456 | 137,459 | $15 |
| 7 | FLUX 3 VideoBfl | 1,449 | 37,042 | $17 |
| 8 | HappyHorse 1.0Aorizon | 1,442 | 70,640 | - |
| 9 | Wan 2.7Wan | 1,428 | 85,521 | - |
| 11 | Grok Imagine VideoSpaceXSI (formerly xAI) | 1,415 | 554,050 | - |
| 12 | Veo 3.1 AudioGoogle | 1,398 | 25,116 | - |
Source: LMArena Image-to-Video Arena, data from September 22, 2026, 2,081,953 votes in this arena. Left out until they have enough votes (LMArena still marks them preliminary): HiDream-O1-Video-1.0.
AI benchmark leaderboard
Fixed tests with right answers, run the same way for every model. Tap a column to sort.| # | Model | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 57.6 | 61.4% | - | 59.6% | 93.3% | 97 t/s | $5.98 | $4 / $20 |
| 2 | Claude Sonnet 5.5Anthropic | 56.0 | 55.0% | - | 63.6% | - | 141 t/s | $5.46 | $2 / $10 |
| 3 | Claude Fable 5.1Anthropic | 53.4 | 59.1% | 93.7% | 52.0% | 90.0% | 71 t/s | $7.63 | $10 / $50 |
| 4 | GPT-6 AstraOpenAI | 52.7 | 54.7% | 96.1% | 59.1% | 95.0% | 57 t/s | $3.26 | $10 / $50 |
| 5 | Gemini 4 ArgonGoogle | 52.6 | 57.1% | - | 57.1% | 91.3% | - | $1.99 | $2 / $10 |
| 6 | Muse WonderMeta | 52.4 | 50.1% | - | 52.0% | - | 181 t/s | $4.52 | $4 / $20 |
| 7 | GPT-6.1 SolOpenAI | 51.8 | 52.9% | - | 56.1% | 94.2% | 66 t/s | $0.72 | $2 / $10 |
| 8 | Muse Spark 1.3Meta | 48.1 | 48.7% | 93.5% | 33.3% | - | 134 t/s | $1.60 | $1.25 / $4.25 |
| 9 | Grok 4.7SpaceXSI (formerly xAI) | 46.4 | 43.1% | - | 25.8% | 61.4% | 93 t/s | $3.74 | $2 / $6 |
| 10 | MiMo-V2.6-ProXiaomiOpen weights | 46.3 | 49.4% | - | 34.8% | - | 48 t/s | $0.13 | $0.435 / $0.87 |
| 11 | Qwen3.8 MaxAlibaba (Qwen) | 45.4 | 43.1% | 92.8% | 38.9% | - | 38 t/s | $5.41 | $2 / $6 |
| 12 | GLM-5.3Z.ai (Zhipu)Open weights | 44.8 | 42.3% | 91.7% | 41.9% | - | 81 t/s | $2.01 | $1.4 / $4.4 |
| 13 | Kimi K3Moonshot AIOpen weights | 43.6 | 46.9% | 93.5% | 12.6% | 60.4% | 42 t/s | $2.00 | $3 / $15 |
| 14 | Claude Haiku 5.5Anthropic | 43.4 | 44.4% | - | 32.8% | 61.3% | 235 t/s | $0.21 | $0.1 / $0.5 |
| 15 | GLM-5.3-FlashZ.ai (Zhipu)Open weights | 41.8 | 39.9% | 91.2% | 32.8% | 65.8% | 71 t/s | $0.28 | $0.15 / $0.5 |
| 16 | Gemini 3.8 FlashGoogle | 40.9 | 47.8% | 95.3% | 19.7% | 89.2% | 147 t/s | $1.24 | $0.75 / $3.75 |
| 17 | DeepSeek V4.1 FlashDeepSeekOpen weights | 39.5 | 39.2% | - | 26.8% | 72.9% | 176 t/s | $0.27 | $0.3 / $1.2 |
| 18 | Muse Spark OpenMetaOpen weights | 38.9 | 43.9% | 89.0% | 8.1% | - | 131 t/s | $0.84 | $1.25 / $4.25 |
| 19 | Mistral Large 4Mistral AI | 38.4 | 35.0% | - | 26.8% | - | 71 t/s | $1.13 | $1.36 / $4.18 |
| 20 | GPT-6 LunaOpenAI | 38.1 | 38.5% | - | 12.6% | 59.3% | 120 t/s | $0.07 | $0.1 / $0.5 |
| 21 | DeepSeek V4 ProDeepSeekOpen weights | 36.0 | 41.0% | 92.8% | 14.1% | 61.3% | 79 t/s | $0.67 | $1.32 / $3.96 |
| 22 | Qwen3.8 27BAlibaba (Qwen)Open weights | 33.7 | 33.9% | 90.5% | 5.6% | 42.4% | 41 t/s | $1.01 | $0.5 / $3 |
| 23 | MiniMax-M3MiniMaxOpen weights | 29.2 | 39.0% | 92.9% | 2.0% | - | 93 t/s | $0.51 | $0.3 / $1.2 |
| 24 | Gemma 4 31BGoogleOpen weights | 14.7 | 23.6% | 85.7% | 0.0% | - | 35 t/s | $0.00 | - |
- Artificial Analysis Intelligence Index
- a combined score across its test suite
- Humanity's Last Exam
- expert-level questions across dozens of fields
- GPQA Diamond
- graduate-level science questions
- Terminal-Bench 4.0
- real tasks done in a computer terminal (Artificial Analysis run)
- ARC-AGI-2
- visual puzzles that test reasoning on problems the model has not seen
- Speed
- output tokens per second on the model's main API
- Cost per task
- what one Artificial Analysis test task costs to run, lower is better
- API price
- input / output price per million tokens, sorted by the output price
How these rankings work
People's choice comes from LMArena, where people ask a question, get two anonymous answers and vote for the better one. With enough votes, that shows which AI people actually prefer for each kind of work. Benchmarks are fixed tests with right answers, run the same way for every model. They are useful for comparing raw ability and price, but they do not always match what feels best in daily use.
Sources
- LMArena (arena.ai) leaderboards, official data export, data from text 2026-10-15; webdev and agent 2026-10-16; vision 2026-10-09; text-to-image 2026-10-07; image edit 2026-10-06; text-to-video and image-to-video 2026-09-22
- Artificial Analysis LLM leaderboard and model pages, data from 2026-10-18
- Artificial Analysis Coding Agent Index v1.5, data from 2026-10-18
- Artificial Analysis image and video arenas, data from 2026-10-18
- Terminal-Bench 4.0 leaderboard (Stanford, Harbor, Laude Institute), data from 2026-10-18
- ARC Prize leaderboard (ARC-AGI-2 and ARC-AGI-3 data files), data from 2026-10-15
- Scale AI Humanity's Last Exam leaderboard, data from 2026-10-18
- Official model and pricing pages (Anthropic, OpenAI, Google, SpaceXSI/xAI, Meta, Alibaba Cloud, Mistral, Moonshot, Z.ai, DeepSeek, Xiaomi, MiniMax, Microsoft AI, ByteDance Seed / BytePlus, Black Forest Labs), data from 2026-10-18
LMArena data is published under the CC BY 4.0 license; the people's scores on this page come from its official data export. Prices change often, so confirm on the provider's pricing page before you commit.
Frequently asked questions
What is the best AI model right now?
In blind votes on LMArena, where people compare two anonymous answers to the same question, Muse Wonder, Claude Opus 5.5, Claude Fable 5, Claude Opus 4.6, Claude Fable 5.1 and Claude Opus 4.7 share first place (a statistical tie). The right pick still depends on the job, so check the row for what you do.
Which AI is best for coding?
People voting on working apps built from the same prompt put Claude Opus 5.5 first. For building sites and tools without coding yourself, that is the model to use in Claude Code or a similar agent.
Which AI is best for writing?
For stories, scripts and copy, Claude Opus 5.5, Claude Fable 5 and Claude Opus 4.6 lead the blind creative-writing votes (a statistical tie).
What is the best AI image generator?
GPT-Image-2 leads the blind text-to-image votes, where people pick the better of two images made from the same prompt.
What is the best AI video generator?
In blind text-to-video votes, Gemini Omni 1.1 Flash and Gemini Omni Flash share first place (a statistical tie).
What is the difference between people's choice and benchmarks?
People's choice comes from blind votes: people see two anonymous answers and pick the one they prefer, so it measures what feels best to use. Benchmarks are fixed tests with right answers, so they measure what a model gets right under the same conditions. The two often disagree, which is why they sit in separate tabs.
How is this leaderboard put together?
Every number comes from a named public source: LMArena's blind votes for the people's choice, and Artificial Analysis, Terminal-Bench and ARC Prize for the benchmarks, with prices from official pricing pages. This edition uses data published between October 13, 2026 and October 18, 2026. Models whose scores overlap within the margin of error share a rank.
New to these tools? Start with the guide what Claude Code is or browse the Claude skills directory.