Leaderboard Changelog

GPT 5.6 Sol (xHigh) has been added to the Agent Arena leaderboard.
Grok 4.5 has been added to the Agent Arena leaderboard.

The claude-fable-5 scores have been updated to reflect only votes collected on or after July 1 for the Code, Vision, Document and Search Arena leaderboards.

seedream-5.0-pro has been added to the Image Arena leaderboard.

muse-spark-1.1 has been added to the Text Arena & Code Arena leaderboards.
gpt-5.6-sol-xhigh (codex-harness) has been added to the Code Arena leaderboard.

reve-2.1 has been added to the Text-to-Image leaderboard.

muse-image has been added to the Text-to-Image & Image Edit leaderboard.
muse-video has been added to the Text-to-Video leaderboard.

Claude Opus 4.7 (Thinking) has been added to the Agent Arena leaderboard.

krea-2-turbo has been added to the Text-to-Image leaderboard.
claude-sonnet-5-thinking has been added to Code, Text, Search, Vision, and Document leaderboards.

gemini-3.1-flash-lite-image (nano-banana-2-lite) has been added to the Text-to-Image and Image Edit leaderboards.
gemini-omni-flash has been added to the Video Edit leaderboard.

qwen-image-2.0-pro-2026-06-22 has been added to the Text-to-Image and Image Edit leaderboards.

qwen3.7-plus-preview has been deprecated from the Vision leaderboard.
qwen3.7-plus has been added to the Vision leaderboard.

wan2.7-i2v has been added to the Image-to-Video leaderboard.

seed-2.1-pro-preview has been added to the Code Arena leaderboard.

muse-spark has been deprecated from the Code Arena leaderboard.

Kimi K2.7 Code has been added to the Agent Arena leaderboard.

Minimax M3 and GLM 5.2 (Max) has been added to the Agent Arena leaderboard. GLM 5.2 (Max) has been added to Text and Code leaderboards.

claude-opus-4-8 and claude-fable-5 has been added to the Search leaderboard.
kimi-k2.7-code has been added to the Code leaderboard.

Nemotron 3 Ultra has been added to the Agent Arena leaderboard.

gemini-omni-flash has been added to the Text-to-Video leaderboard.
GPT 5.5 (xHigh) has been added to the Agent Arena leaderboard.

claude-fable-5 has been added to the Code, Text, Document, Vision, and Agent Arena leaderboard.

Claude Opus 4.8 (Thinking) & Claude Opus 4.8 has been added to the Agent Arena leaderboard.

Grok Build 0.1 & Grok 4.3 (High) has been added to the Agent Arena leaderboard.

claude-opus-4-8, and claude-opus-4-8-thinking has been added to the Text, Vision, Document, and Code leaderboards.

krea-2-medium, krea-2-large, Cosmos3-Super-Text2Image, and Cosmos3-Super-Text2Image (Agentic) has been added to the Text-to-Image leaderboard.

mistral-medium-3.5 has been added to the Code leaderboard.

The Agent Arena leaderboard is now live. It ranks models on real-world agentic evals at scale, measuring behavioral signals like file downloads, disapproval events, retries, and steerability rather than static benchmark scores or preference votes alone. Check out the Agent Arena blog post to learn more about the methodology.

nvidia-nemotron-3-ultra-550b-a55b-nvfp4 has been added to the Text-to-Image leaderboard.

mistral-medium-3.5 has been added to the Text and Vision leaderboards.

ideogram-4.0-quality has been added to the Text-to-Image leaderboard.
reve-2.0 has been added to the Text-to-Image and Image Edit leaderboard.

mai-image-2.5 has been added to the Image Edit leaderboard.

wan2.7-t2v has been added to the Text-to-Video leaderboard.

mai-image-2.5-preview has been added to the Text-to-Image leaderboard.

qwen3.7-max-20260517 has been added to the Code leaderboard.

recraft-v4.1-pro and recraft-v4.1-utility-pro has been added to the Text-to-Image leaderboard.

hidream-o1-image has been added to the Text-to-Image leaderboard.
grok-imagine-image-quality (20260519) has been added to the Image Edit leaderboard.

gemini-3.5-flash has been added to Text and Code leaderboards.

qwen3.7-max-preview has been added to the Text and Vision leaderboards.
granite-4.1-8b has been added to the Text and Code leaderboards.

trinity-large-thinking has been added to the Text leaderboard.
gpt-5.5-xhigh (codex-harness) has been added to the Code leaderboard.

Votes collected from Battles in Direct now count toward leaderboards.

Since March 2026, we've been running an experiment where 10% of direct-chat sessions on Arena are converted into battles between two randomly sampled models. Starting today, votes collected this way will be included in leaderboards. Over the next month, we'll also gradually backfill Battles in Direct votes collected from March through May, so models active during that time period may see vote counts increase and scores shift slightly.

This change meaningfully increases daily vote volume, which tightens confidence intervals and stabilizes ranks faster. The distribution of prompts also shifts: direct battles arrive with prior conversational context and skew toward longer queries, harder prompts, coding, and instruction following. Direct battles where the user votes skip are not used, so every direct-battle vote is a decisive A-or-B preference.

We observed two new biases in the Battles in Direct voting data and corrected for them in the Bradley-Terry fit for the text, vision, search, and document arenas. The first is a position bias favoring Model A (the model on the left); the second is an advantage given to models that share an organization with the prior turns of context. To correct each bias, we added a corresponding feature to the model: is_direct_battle and same_org_indicator, respectively. The model fitting learns coefficients for these features that absorb the biases.

gpt-5.5-instant has been added to the Text, Vision, and Document leaderboard.
ernie-5.1 has been added to Search leaderboard.

gemma-4-31b and gemma-4-26b-a4b has been added to the Vision leaderboard.
qwen3.6-max-preview and hunyuan-hy3-preview has been added to Code leaderboard.

grok-imagine-image-quality has been added to the Text-to-Image leaderboard and Image edit leaderboard.
gpt-5.5-instant has been added to the Vision, Text and Document leaderboards.

uni-1.1-max and uni-1.1 has been added to the Text-to-Image leaderboard and Image Edit leaderboard.
gpt-image-2 (medium) has been updated on the Image leaderboards. The updated score reflects performance across Arena's full user base. As public usage scales, scores stabilize to reflect a larger sample of real-world use.
gemma-4-26b-a4b & gemma-4-31b have been added to the Code Arena leaderboard.

mimo-v2-omni has been added to the Vision leaderboard.
grok-4.3 has been added to the Text, Search, Code and Vision leaderboards.
trinity-large-thinking has been added to the Text and Code leaderboards.

qwen3.6-max-preview has been added to the Text leaderboard.
hunyuan-hy3-preview has been added to the Text leaderboard.
glm-5v-turbo has been added to the Vision leaderboard.

ernie-5.1-preview has been added to the Text leaderboard.

gpt-5.5-high has been added to Code, Text, Expert, Search, Document, and Vision leaderboards.
claude-opus-4-7 has been added to the Search leaderboard.

qwen-image-2.0-pro-2026-04-22 has been added to the Text-to-Image and Image Edit leaderboards.
deepseek-v4-pro, deepseek-v4-pro-thinking, and deepseek-v4-flash-thinking has been added to the Text and Code leaderboards.

kimi-k2.6 has been added to Code, Vision, Document, and Text leaderboards.
muse-spark has been added to Code Arena leaderboard.

claude-opus-4-7-thinking has been added to the Code leaderboard.
dreamina-seedance-2.0-720p has been added to the Video Edit leaderboard.
gpt-image-2 (medium) has been added to the Text-to-Image and Image Edit leaderboard.

claude-opus-4-6-thinking and claude-opus-4-7 have been added to the Vision and Document leaderboards.
qwen3.6-plus has been added to the Text and Code leaderboards.

kling-image-o1 has been added to the Image Edit leaderboard
claude-opus-4-6-thinking and claude-opus-4-7 have been added to the Text and Code Arena WebDev leaderboards

The Image to WebDev leaderboard is now live.

happyhorse-1.0 is live on the Video Edit leaderboard.
claude-opus-4-6-thinking, kimi-k2.5-thinking, gemma-4-31b and grok-4.20-beta-0309-reasoning have been added to the Document Arena leaderboard.

trinity-large has been renamed to trinity-large-preview for clarity on the Text leaderboard.
wan2.7-image and wan2.7-image-pro have been added to the Text-to-Image and Image Edit leaderboards.

muse-spark has been added to the Text leaderboard.
glm-5.1 has been added to the Code leaderboard.

glm-5.1 has been added to the Text leaderboard.

dola-seed-2.0-pro has been added to the Text leaderboard.

qwen3.6-plus-preview has been added to the Code leaderboard.
gpt-5.4-search has been added to the Search leaderboard.
gpt-5.4-mini-high has been added to the Code (WebDev) leaderboard
longcat-flash-chat has been added to the Text leaderboard.
gemma-4-31b and gemma-4-26b-a4b have been added to the Text leaderboard.

claude-opus-4-6, claude-opus-4-6-thinking and claude-sonnet-4-6 have been added to the Vision leaderboard.
grok-4.20-multi-agent-beta-0309 is now on the Text, Vision, and Search leaderboards.

gemini-3-flash-grounding has been added to the Search leaderboard.
mercury-2 has been added to the Text and Code Area leaderboards.
runway-gen4 has been added to the Text-to-Image leaderboard.
gpt-5.4-nano-high and gpt-5.4-mini-high have been added to the Text leaderboard.

mimo-v2-pro has been added to the Text and Code leaderboards.

qwen3.5-max-preview has been added to the Text leaderboard.
mai-image-2 has been added to the Text-to-Image leaderboard.
grok-4.20-beta-0309-reasoning has been added to the Vision leaderboard.

minimax-m2.7 has been added to the Text and Code leaderboards.

grok-4.20-beta-0309-reasoning has been added to the Text and Code leaderboards.
amazon-nova-experimental-chat-26-02-10 has been added to the Text leaderboard.

gpt-5.4-high (codex-harness), gpt-5.4-medium (codex-harness), and gpt-5.3-codex (codex-harness) have been added to the Code leaderboards.

nvidia-nemotron-3-super-120b-a12b has been added to the Text leaderboard.
gpt-5.4 is on the Text and Document leaderboards.

The Video Arena leaderboards will reflect a single consolidated vote category going forward.

Previously, prompts and generations were visible to a broader Discord community, allowing non-authors to vote. After the migration to arena.ai, voting is now limited to the prompt author only. We are discontinuing the separate “Author Vote” category on the Video Arena leaderboards as all votes now originate from prompt authors.

pixverse-v5.6 has been added to the Text-to-Video and Image-to-Video leaderboards

gpt-5.4-high and gpt-5.4 are on the Text leaderboard

qwen3.5-flash has been added to the Text and Code leaderboards.

gemini-3.1-flash-lite-preview has been added to the Text and Code leaderboards.
qwen3.5-122b-a10b, qwen3.5-35b-a3b and qwen3.5-27b have been added to the Text and Code leaderboards.

The Document Arena leaderboard has been added. The Document Arena displays model rankings based on side-by-side evaluations of real-world document reasoning performance across user-uploaded PDF files.

runway-gen-4.5 has been added to the Text-to-Video leaderboard

An older version of grok-imagine-image has been renamed to grok-imagine-image (20260207) in Image Edit to reflect the dated snapshot. The current default model now listed as grok-imagine-image represents the latest version.

p-video has been added to the Text-to-Video and Image-to-Video leaderboards.
gemini-3.1-flash-image-preview (nano-banana-2) [web-search] has been added to the Text-to-Image and Image Edit leaderboards.
claude-opus-4-6-search has been added to the Search leaderboard.
kling-v3-pro has been added to the Image-to-Video leaderboard.

Code Arena WebDev Leaderboard Categorization
The Code Arena WebDev leaderboard is now segmented into two distinct categories to reflect different generation environments and capability surfaces:

HTML: ranks standalone HTML/CSS/JS generation within a single document scope.React: ranks multi-file application generation, including component structure, routing, state coordination, and cross-file consistency.

This separation provides clearer visibility into model performance across simple document-level outputs vs. structured application-level builds.

gpt-5.2-chat-latest-20260210 has been added to the Vision leaderboard.
grok-4.20-beta1 has been added to the Text and Search leaderboards.
seedream-5.0-lite has been added to the Text-to-Image and Image Edit leaderboards.

qwen3.5-397b-a17b has been added to the Code leaderboard.
wan2.6-i2v and wan2.6-t2v have been added to the Video leaderboard
amazon-nova-experimental-chat-26-01-10 has been added to the Text leaderboard.

trinity-large has been added to the Text leaderboard.
reve-v1.5 has been added to the Text-to-Image leaderboard.

gpt-5.2-chat-latest-20260210 has been added to the Text leaderboard.

claude-sonnet-4-6 has been added to the Text and Code leaderboard.
qwen3.5-397b-a17b has been added to the Vision leaderboard.
recraft-v4 has been added to the Text-to-Image leaderboard.

molmo-2-8b has been added to the Vision and leaderboard.
gemini-3.1-pro-preview has been added to the Text, Vision and Code leaderboards.
mimo-v2-flash (thinking) has been added to the Text and Code leaderboards.

minimax-m2.5 has been added to the Text and Code leaderboards.

dola-seed-2.0-preview has been added to the Text and Vision leaderboards.

glm-5 has been added to the Code leaderboard.

glm-5 has been added to the Text leaderboard.

veo-3.1-fast-audio-1080p and veo-3.1-audio-1080p have been added to the Text-to-Video and Image-to-Video leaderboards.

step-3.5-flash has been added to the Text leaderboard.

claude-opus-4-6-thinking has been added to the Text and Code leaderboards.

We’ve updated the Text-to-Image Arena with:

Prompt Categories: category-specific leaderboards for clearer domain-level performance Quality Filtering: reducing noisy or underspecified prompts for more reliable rankings

Older categories for user-generated vs. pre-generated prompts have been deprecated.

grok-imagine-image-pro and grok-imagine-image have been added to the Text-to-Image and Image Edit leaderboards.

claude-opus-4-6 has been added to the Text and Code leaderboards.
kimi-k2.5-instant has been added to the Text, Vision and Code leaderboards.

vidu-q3-pro is on the Image-to-Video leaderboard

grok-imagine-video-720p is on the Image-to-Video leaderboard

kimi-k2.5-thinking is on the Code Arena, WebDev leaderboard

glm-4.7-flash has been added to the Text leaderboard.
devstral-2 has been added to the Code leaderboard.

The following models have been added to the Text-to-Video leaderboard:
kling-o1-pro
ltx-2-19b

ltx-2-19b has been added to the Image-to-Video leaderboard

The following models have been added to the Image Edit leaderboard:
p-image-edit
wan2.6-image

The following models have been added to the Text-to-Image leaderboard:
p-image
wan2.6-t2i

The following models have been added to the Search leaderboard:
gemini-3-flash-grounding
claude-sonnet-4-5-search
claude-opus-4-5-search
gpt-5.2-search-non-reasoning

kimi-k2.5-thinking has been added to the Vision leaderboard

grok-imagine-video-480p is the Text-to-Video and Image-to-Video leaderboards

kimi-k2.5-thinking is on the Text Leaderboard.

hunyuan-image-3.0-instruct is on the Image Edit leaderboard

We've added Multi-Image Edit as new category to the Image Edit leaderboard which consists of votes where multiple images were passed into the models. The previous overall leaderboard has been renamed to Single-Image Edit.

gpt-5.2-codex has been added to the Code leaderboard.

Video Arena leaderboards now include votes collected from both the lmarena.ai website and our Discord server. Read our blog for more details.

olmo-3.1-32b-instruct and olmo-3.1-32b-think have been added to the Text leaderboard.
glm-image was added to the Text-to-Image leaderboard.

wan2.5-i2i-preview has been added to the Image Edit leaderboard

glm-4.6v has been added to the Vision leaderboard.
gemini-3-flash (thinking-minimal) has been updated on the Vision leaderboard.
flux-2-klein-9b & flux-2-klein-4b have been added to the Text-to-Image and Image Edit leaderboards.
z-image-turbo has been added to the Text-to-Image leaderboard.

ernie-5.0-0110 has been added to the Text leaderboard.

We’ve completed a major improvement to our data pipeline that resolves several known issues and applies data filtering more consistently:

The validation process led to minimal adjustments in leaderboard rankings. Models and leaderboards with fewer votes may see larger score fluctuations.

Here's a summary of the changes in the new pipeline:

Vote filters, such as identity leak detection and data quality filtering, are now applied more consistently across all votes.Vote de-duplication is now enabled in text-to-image and video arenas.

hunyuan-video-1.5 has been added to the Text-to-Video and Image-to-Video leaderboards

ernie-5.0-preview-1220 has been added to the Vision leaderboard
seedance-v1.5-pro has been added to the Text-to-Video and Image-to-Video leaderboards.

minimax-m2.1-preview and glm-4.7 have been added to the Text leaderboard.

minimax-m2.1-preview has been added to the WebDev leaderboard.

mimo-v2-flash (non-thinking) has been added to the Text and WebDev leaderboards.

ernie-5.0-preview-1203 has been added to the Text leaderboard.
glm-4.7 has been added to the WebDev leaderboard powered by the new Code Arena.

amazon-nova-experimental-chat-11-10 has been added to the Text leaderboard.

gpt-5.2 has been added to the Text leaderboard.
grok-4-1-fast-search and gpt-5.2-search are on the Search leaderboard.
reve-v1.1 and reve-v1.1-fast are on the Image Edit leaderboard.

gemini-3-flash and gemini-3-flash (thinking-minimal) have been added to the Text, Vision and WebDev leaderboards.

flux-2-max is on the Text-to-Image and Image Edit leaderboards.
gpt-image-1.5 is on the Text-to-Image and Image Edit leaderboards.
chatgpt-image-latest (20251216) has been added to the Image Edit leaderboard.
gpt-5.2-high has been added to the Text leaderboard.
ibm-granite-h-small has been added to the Text leaderboard.

nvidia-nemotron-3-nano-30b-a3b-bf16 has been added to the Text leaderboard

kandinsky-5.0-t2v-lite and kandinsky-5.0-t2v-pro have been added to the Text-to-Video.
kling-2.6-pro has been added to the Text-to-Video and Image-to-Video leaderboards.
flux-2-dev has been added to the Text-to-Image and Image Edit leaderboards.

gpt-5.2 and gpt-5.2-high have been added to the WebDev leaderboard powered by Code Arena.

mistral-large-3 has been added to the WebDev leaderboard powered by Code Arena.
intellect-3 has been added to the Text leaderboard.

ernie-5.0-preview-1103 and nova-2-lite have been added to the Text leaderboard.

grok-4-fast has been renamed to grok-4-fast-chat to better reflect the specific model variant. Additionally, olmo-3-32b-think has been added to the Text leaderboard.

grok-4-fast-reasoning has been added to the Text leaderboard.
devstral-medium-2507 has been added to the new WebDev leaderboard (powered by Code Arena).
gpt-5.1-high and gpt-5.1 has been added to the Vision leaderboard.
deepseek-v3.2 and deepseek-v3.2-thinking have been added to the Text leaderboard.
seedream-4.5 has been added to the Image Edit and Text-to-Image leaderboards

gpt-5.1-search and gemini-3-pro-grounding have been added to the Search leaderboard.

mistral-large-3 has been added to the Text leaderboard.
wan2.5-t2v-preview has been added to the Text-to-Video leaderboard.
gemini-3-pro-image-preview-2k (nano-banana-pro) has been added to the Text-to-Image and Image Edit leaderboards.

KAT-Coder-Pro-V1 has been added to the new WebDev leaderboard.
flux-2-flex and flux-2-pro have been added to the Text-to-Image and Image Edit leaderboards.

claude-opus-4-5-20251101 and claude-opus-4-5-20251101-thinking-32k have been added to the Text and WebDev leaderboards

ernie-5.0-preview-1120 has been added to the Vision leaderboard.
gemini-3-pro-image-preview (nano-banana-pro) has been added to the Text-to-Image and Image Edit leaderboards.

Additionally, the following models have been added to the new WebDev leaderboard:

gpt-5.1gpt-5.1-mediumgpt-5.1-codexgpt-5.1-codex-mini

mercury has been added to the Text leaderboard.

cogito v2.1 has been added to the WebDev leaderboard.
gpt-5.1-high and gpt-5.1 have been added to the Text leaderboard.

gemini-3-pro has been added to the Text, Vision and WebDev leaderboards.

grok-4.1-thinking and grok-4.1 have been added to the Text leaderboard.
wan2.5-t2i-preview has been added to the Text-to-Image leaderboard.
wan2.5-i2v-preview has been added to the Image-to-Video leaderboard.

ray-3 has been added to the Text-to-Video and Image-to-Video leaderboards.

vidu-q2-turbo & vidu-q2-pro are now on the Image-to-Video leaderboard.

The WebDev leaderboard is now powered by the Code Arena experience.
amazon-nova-experimental-chat-10-20 has been added to the Text leaderboard.

ernie-5.0-preview-1022 has been added to the Text leaderboard.
reve-edit-fast has been added to the Image Edit leaderboard.

gpt-image-1-mini has been added to the Text-to-Image and Image Edit leaderboards.

Introducing Arena Expert: a new LMArena evaluation framework to identify the toughest, most expert-level prompts from real users, powering a new Expert leaderboard.

We also introduce Occupational Categories that underlie eight new leaderboards:

Software & IT Services Writing, Literature, & LanguageLife, Physical, & Social ScienceEntertainment, Sports, & MediaBusiness, Management, & Financial OpsMathematicalLegal & GovernmentMedicine & Healthcare

Arena Expert aims to sharpen the difficulty level compared to Arena Hard. While Hard includes about a third of all LMArena prompts, Arena Expert includes only 5.5% of all prompts. Expert prompts are identified by their reasoning depth and specificity, producing sharper separations between models. By mapping all Arena prompts across occupational fields, the Occupational Categories system captures the full spectrum of real-world reasoning tasks.

→ Read more on our blog: http://news.lmarena.ai/arena-expert

MiniMax-M2 has been added to the WebDev leaderboard.

Hailuo 2.3 has been added to the Text-to-Video leaderboard.

Hailuo 2.3 has been added to the Image-to-Video leaderboard.

The following models have been added to the WebDev leaderboard:

Claude-Haiku-4-5-20251001Qwen3-235b-a22b-instruct-2507Claude-sonnet-4-5-20250929-thinking-32kGLM-4.6

Additionally, Veo 3.1 variants have been added to the Text-to-Video and Image-to-Video leaderboards.

Claude Haiku 4.5 and Amazon-Nova-Experimental-Chat-10-09 have been added to the Text leaderboard.

We've also refined the logic for our Coding category to improve precision. Prompts that resembled code, but are not coding related (such as markdown) have been removed. The new rule has been applied retroactively on data, so while the Coding category is now smaller, it’s more accurate.

Sora 2 and Sora 2 Pro has been added to the Text-to-Video leaderboard.

MAI-1-Image has been added to the Text-to-Image leaderboard.
Kling 2.5 Turbo 1080p has been added to the Text-to-Video and Image-to-Video leaderboards.

DeepSeek-V3.2-Exp and the thinking variant have been added to the Text leaderboard.

Ling Flash 2.0 and Ring Flash 2.0 have been added to the Text leaderboard.

Hunyuan Vision 1.5 Thinking has been added to the Vision leaderboard.

Hunyuan Image 3.0 has been added to the Text-to-Image leaderboard.

We added a filter to remove rows where a model battles against itself. This happens very rarely, in instances where we briefly serve the same model from two different API endpoints at the same time.

Claude Sonnet 4.5 Thinking 32k and GLM 4.6 have been added to the Text leaderboard.

Claude Sonnet 4.5 has been added to the Text and Web Dev leaderboards.

Reve V1 has been added to the Image Edit leaderboard.

The following models have been added to the Text leaderboard:

deepseek-v3.1-terminusdeepseek-v3.1-terminus-thinkinggemini-2.5-flash-lite-preview-09-2025gemini-2.5-flash-preview-09-2025qwen3-max-2025-09-23qwen3-vl-235b-a22b-instructqwen3-vl-235b-a22b-thinking

New model announcement:
Seedream-4-2k has been added to the Text-to-Image and Image Edit leaderboards.

Note that Seedream-4-high-res-fal and Seedream-4-fal are variants run on the fal.ai platform. Due to differences in hosting, they are named separately as distinct models. Seedream-4-2k is the official endpoint provided by ByteDance.

New model announcements:
Grok-4-fast has been added to the Text leaderboard.
Grok-4-fast-search has been added to the Search leaderboard.

We have added a new "preliminary" tag to the leaderboard. If a model is tested anonymously and is subsequently released publicly, we mark its score as "preliminary" until enough fresh votes have been collected after the model’s public release. The tag indicates that scores may shift as community prompts and votes evolve after public launch. See our leaderboard policy for more details about evaluating models.

New model announcements:
Longcat-Flash-Chat, Qwen 3 Next-80b-a3b-instruct and Qwen 3 Next-80b-a3b-thinking have been added to the Text leaderboards.

We've updated our data pipeline to add a filter which removes votes from users who exhibit statistically anomalous voting patterns. This improves the quality of the rankings by removing votes from users whose votes are arbitrary, rather than based on the quality of the responses.

New model announcement:
Seedream 4 High Res has been added to the Text-to-Image and Image Edit leaderboards.

Deepseek v3.1 & Deepseek v3.1-thinking have been added to the WebDev leaderboard.

New model announcement:
Seedream 4 has been added to the Text-to-Image and Image Edit leaderboards.

New model announcements:
Qwen3-max-preview and Kimi-K2-0905-preview have been added to the Text Leaderboard.

We also enabled filtering for the mistaken image generation and image edit requests for text arena.

Due to the increase in image generation traffic brought by nano-banana, we noticed there were prompts in our vision arena data which were asking for image generation but did not have image output enabled. We've implemented an LLM based rule to filter these rows out from the vision leaderboard calculation.

New model announcements:
Diffbot-small-xl has been added to the Search Leaderboard
Qwen-3-Image-Prompt-Extend has been added to the Text-to-Image Leaderboard.

The following have been added to the Text Leaderboard:

DeepSeek V3.1 (thinking and non-thinking)Hunyuan-t1-20250711

New model announcement: MAI-1-preview has been added to the Text Leaderboard.

New model announcement: Gemini-2.5-Flash-Image-Preview ("nano-banana") has been added to the Text-to-Image and Image Edit leaderboards.

GPT-5 and Claude Opus 4.1 have been added to the Search Leaderboard.

We also noticed that the effects of response style are different for search than they are for ordinary chat so we'll be defaulting this leaderboard to ordinary Bradley-Terry while we study search specific style effects.

New model announcements
The following have been added to the Text and Vision leaderboards:

Qwen-vl-max-2025Step 3 Mistral-Medium-2508 GLM-4.5V

Lucid Origin has been added to the Text-to-Image leaderboard.
Ray 2 has been added to the Text-to-Video and Image-to-Video leaderboards
Runway Gen 4 Turbo has been added to the Image-to-Video leaderboard

New model announcement: Qwen-Image-Edit has been added to the Image Edit leaderboard.

New model announcement: Claude Opus 4.1 Thinking as been added to the Text and WebDev Arena leaderboards. Sora has been added to the Text-to-Image leaderboard.

New model announcement: three additional gpt-5 models are on the Text Leaderboard. These three reasoning models were configured with the highest reasoning setting.

gpt-5-chatgpt-5-mini-highgpt-5-nano-high

New model announcement: gpt-0ss-120b and gpt-oss-20b have been added to the Text and WebDev leaderboards. Hailuo 2 Pro versions have been added to the Text-to-Video and Image-to-Video leaderboards.

New model announcement: Claude Opus 4.1 is on Text and WebDev leaderboards.

New model announcement: GPT-5 is on the Text, WebDev, and Vision leaderboards.

Big update: three new leaderboards!
Check out the Search, Text-to-Video, and Image-to-Video leaderboards.

Since the video arenas are used through our discord server, there are a few considerations we made for handling the votes. Currently, the model identities are revealed after two votes are cast on a generation. For fairness, we only use the votes cast before the model names are revealed when constructing the leaderboard.

The video arenas are also the first arenas where multiple votes can be cast on the same pair of generations, so unlike the other arenas, some votes are cast by people other than the author of the prompt. The overall leaderboard is computed using all anonymous votes, and we've created a new category which uses only the votes cast by the prompt's author.

We have updated the "total votes" counts to include battles involving models not included on the leaderboard (for example, due to being deprecated). The battles between these models and models present on the leaderboard are informative of model strengths, even if the former are not shown, and thus help reduce the variance of the scores. The leaderboard computation is not changing; you will only see a change in the vote counts.

New model announcements: GLM-4.5 and GLM-4.5 Air are now on the Text leaderboard.

New model announcement: Qwen3-235b-a22b-instruct-2507 is now on the Text leaderboard.

New model announcements: Qwen3-Coder and Kimi K2 are now on the WebDev leaderboard.

New model announcements! Imagen 4 Generate Preview 06-06 v2 and Imagen 4 Ultra Generate Preview 06-06 v2 are now on the Text-to-Image leaderboard.

We made improvements to the methodology behind Arena scores!

Our leaderboard uses confidence intervals to represent the uncertainty and variability inherent in estimating scores based on human voting. Up until now, our confidence intervals have been computed via bootstrapping, a process where we resample the dataset many times, calculate scores on each, and then look at the distribution of the scores over all the runs. While statistically sound, this is computationally intensive, especially with a large number of battles. We’ve recently moved to a new method based on the Central Limit Theorem (CLT) for M-estimators, which allows us to compute confidence intervals via a closed form equation.

We validated this approach by comparing the confidence intervals computed via bootstrapping, with those using the CLT and confirmed that the results are in very close parity (with a fraction of the compute cost and time!). See below:

On LMArena, every vote counts towards producing the leaderboard, but what happens when some models appear more than others? When new models are released, they inevitably have fewer votes than those which have been in use for a while, and when models are deprecated it becomes impossible to collect more votes for them.

To counteract this imbalance and produce a leaderboard that is fair and equally representative of all models, we use an improved reweighting scheme that reweights battles inversely proportionally to how frequently they appear.

The CLT confidence intervals above take these weights into account. Reweighting increases the variance of Arena scores, and we observe wider confidence intervals as a result. This mean that the new rankings will have more ties due to overlapping confidence intervals, especially when there are fewer votes per model like in the vision arena.

New model announcements! Kimi K2 is on the Text leaderboard, Seededit 3 is on the Image Edit leaderboard, and Grok 4 is on the Vision leaderboard.

We're announcing four new models! Grok 4 is on the Text and WebDev leaderboards, Claude Opus 4 Thinking is on the Text leaderboard, Claude Sonnet 4 Thinking is on the Text leaderboard, and Seedream 3 is on the Text-to-Image leaderboard.

We made improvements to our data processing—in particular, we strengthened our deduplication and identity leak detection pipelines.

Deduplication aims to reduce the impact of over-represented or repetitive conversations using a hash-based approach. We count how many times each unique prompt appears. Prompts in the top 0.5% percentile are considered high-frequency. For these high-frequency prompts, we keep only a limited number of samples and discard the rest. Deduplication filters out around 10% of all submitted votes.

Identity leak detection filters out user prompts whose intent is to reveal model information. We first use an LLM classifier to label conversations as identity_leak if they include user prompts that directly attempt to extract or expose model details (e.g., "What is your name?"). We filter out conversations labeled as identity_leak, as well as associated conversations. Less than 4% of all votes are labeled as identity_leak.

We're excited to continue iterating and improving our data processing pipeline!