Skip to content

LeaderboardsSee the latest model rankings

CompareCompare models side by side

ModelsBrowse the model directory

BenchmarksExplore benchmark evidence

ImageImage model rankings and context

VideoVideo model rankings and context

VoiceVoice model rankings and context

Image ArenaHead-to-head image model voting

Video ArenaHead-to-head video model voting

AI NewsRead the latest AI stories

How Rankings WorkUnderstand the ranking methodology

DataLumina Bench data infrastructure.

ChangelogReview ranking and product updates

ContactContact the Lumina team

AdvertiseReview commercial partnership options

Latest rankings

Navigation

Lumina Bench

LeaderboardsSee the latest model rankings

CompareCompare models side by side

ModelsBrowse the model directory

OpenAIOpenAI models AnthropicAnthropic models GoogleGoogle models xAIxAI models MetaMeta models AlibabaAlibaba models Z.AIZ.AI models MoonshotMoonshot models All modelsBrowse every model

BenchmarksExplore benchmark evidence

ImageImage model rankings and context

VideoVideo model rankings and context

VoiceVoice model rankings and context

Arena

Image ArenaHead-to-head image model voting

Video ArenaHead-to-head video model voting

News

AI NewsRead the latest AI stories

Research

How Rankings WorkUnderstand the ranking methodology DataLumina Bench data infrastructure.ChangelogReview ranking and product updates

Contact

ContactContact the Lumina team AdvertiseReview commercial partnership options

Latest rankings

Independent rankings. Commercial relationships never affect scoring.

Loading table

Loading Lumina Bench

Independent Score rankings from hard benchmarks — speed and price stay decision context, not rank inputs.

Product

Leaderboard
Compare
Models
Benchmarks

Research

How rankings work
Data
Changelog

Company

News
Contact
Advertise

© 2026 Lumina Bench@LuminaBench on X ↗

Lumina Bench
Models
DeepSeek V4 Pro 0813

DeepSeek · stable

DeepSeek V4 Pro 0813

DeepSeek's August 2026 GA V4 Pro revision for agent workflows, with non-thinking and configurable low, high and max thinking modes.

#10· 70.7 · Rank #10 · Confidence C

Loading DeepSeek V4 Pro 0813 profile