LLM AtlasLLM AtlasSearch models

Leaderboard

safety rankings

Updated weekly

Each leaderboard uses transparent weighted scoring, current model context, and supporting analysis to help teams interpret the results with confidence. Only full-profile entries appear in rankings; broader catalog records remain available elsewhere on the site when only source-backed metadata is currently available.

Full-profile entries
245

Models with complete enough metadata and scoring coverage to be meaningfully ranked in this category.

Ranking basis
safety

Scores combine benchmark evidence, product metadata, and cost/context signals when those fields are published.

Catalog caveat
25

Tracked models without full scoring remain in the directory and provider pages, but are not relied on for analytical ranking claims.

RankModelProviderScoreContext
#1Claude Opus 4.6Anthropic921,000,000
#2Claude Sonnet 4.5Anthropic921,000,000
#3Claude Sonnet 4.6Anthropic921,000,000
#4Claude 3.7 SonnetAnthropic91200,000
#5Claude Opus 4.1Anthropic90200,000
#6Claude Sonnet 4Anthropic901,000,000
#7Claude Opus 4Anthropic89200,000
#8Command R+ 2026Cohere89128,000
#9Gemini 2.5 ProGoogle DeepMind881,048,576
#10GPT-4oOpenAI88128,000
#11GPT-5.4OpenAI881,000,000
#12Claude Haiku 4.5Anthropic87200,000
#13Gemini 3.1 ProGoogle DeepMind861,048,576
#14GPT-5.2OpenAI861,000,000
#15GPT-5.4 ProOpenAI861,000,000
#16Gemini 3.0 ProGoogle DeepMind851,048,576
#17Gemini 3.1 FlashGoogle DeepMind851,048,576
#18GPT-5OpenAI851,000,000
#19GPT-5.2 ProOpenAI851,000,000
#20GPT-5.3 InstantOpenAI851,000,000
#21Claude Opus 3Anthropic84200,000
#22Claude Sonnet 3Anthropic84200,000
#23Gemini 2.0 FlashGoogle DeepMind841,048,576
#24Gemini 2.5 FlashGoogle DeepMind841,048,576
#25Gemini 2.5 Pro TTSGoogle DeepMind841,048,576
#26Gemini 3.0 FlashGoogle DeepMind841,048,576
#27GPT-5 miniOpenAI841,000,000
#28GPT-5.3-CodexOpenAI841,000,000
#29Mistral Large 25Mistral AI84128,000
#30Sonar Deep ResearchPerplexity84200,000
#31Claude Haiku 3.5Anthropic83200,000
#32Gemini 2.5 Flash LiveGoogle DeepMind831,048,576
#33Gemini 2.5 Flash Native Audio PreviewGoogle DeepMind831,048,576
#34Gemini 2.5 Flash-LiteGoogle DeepMind831,048,576
#35Gemini 3.1 Flash-LiteGoogle DeepMind831,048,576
#36GPT-4.1OpenAI831,048,576
#37GPT-4o-miniOpenAI83128,000
#38Llama Guard 4 12BMeta83131,072
#39Nova ProAmazon Web Services83300,000
#40o4-miniOpenAI83200,000
#41Sonar Reasoning ProPerplexity83200,000
#42GPT-4.1 miniOpenAI821,048,576
#43GPT-5 nanoOpenAI821,000,000
#44Sonar ProPerplexity82200,000
#45Claude Haiku 3Anthropic81200,000
#46Gemini 1.5 ProGoogle DeepMind812,097,152
#47Gemini 2.0 Flash-LiteGoogle DeepMind811,048,576
#48Llama 4 MaverickMeta811,048,576
#49Nova LiteAmazon Web Services81300,000
#50o4-mini-deep-researchOpenAI81200,000
#51SonarPerplexity81128,000
#52Gemini 1.5 FlashGoogle DeepMind801,048,576
#53Llama Guard 3 11B VisionMeta80131,072
#54o1-miniOpenAI80128,000
#55o3OpenAI80200,000
#56o3-deep-researchOpenAI80200,000
#57morph-v3-fast-applyMorph79128,000
#58Nova MicroAmazon Web Services79128,000
#59warpgrep-v2Morph79128,000
#60Gemini 1.5 Flash-8BGoogle DeepMind781,048,576
#61gpt-audioOpenAI78128,000
#62gpt-realtimeOpenAI78128,000
#63o1OpenAI78200,000
#64flash-compactMorph77200,000
#65Claude Haiku 3Anthropic76200,000
#66CodestralMistral AI76256,000
#67GPT Image 1OpenAI7632,768
#68gpt-audio-miniOpenAI76128,000
#69gpt-realtime-miniOpenAI76128,000
#70Voxtral Mini TranscribeMistral AI76131,072
#71chatgpt-image-latestOpenAI7532,768
#72Claude Haiku 3.5Anthropic75200,000
#73Claude Haiku 4.5Anthropic75200,000
#74Codestral EmbedMistral AI7532,768
#75CogVideoXZ.AI75128,000
#76CogView 4Z.AI75128,000
#77Devstral Medium 1.0Mistral AI75128,000
#78Doubao-Seed-1.6ByteDance / Doubao75128,000
#79Doubao-Seed-1.6-FlashByteDance / Doubao75128,000
#80Doubao-Seed-2.0-CodeByteDance / Doubao75128,000
#81Doubao-Seed-CodeByteDance / Doubao75128,000
#82ERNIE 3.5 128KBaidu / ERNIE75128,000
#83ERNIE 4.0 Turbo 8KBaidu / ERNIE75128,000
#84ERNIE Functions 8KBaidu / ERNIE75128,000
#85ERNIE Speed 128KBaidu / ERNIE75128,000
#86GLM-4.5Z.AI75128,000
#87GLM-4.5VZ.AI75128,000
#88GLM-4.6Z.AI75128,000
#89GLM-4.6VZ.AI75128,000
#90GLM-4.7Z.AI75128,000
#91GLM-5Z.AI75128,000
#92GLM-ImageZ.AI75128,000
#93GLM-OCRZ.AI75128,000
#94gpt-image-1-miniOpenAI7532,768
#95Hunyuan CodeTencent / Hunyuan75128,000
#96Hunyuan LiteTencent / Hunyuan75128,000
#97Hunyuan StandardTencent / Hunyuan75128,000
#98Hunyuan T1Tencent / Hunyuan75256,000
#99Hunyuan T1 VisionTencent / Hunyuan75128,000
#100Hunyuan TurboSTencent / Hunyuan75128,000
#101Hunyuan TurboS LongText 128KTencent / Hunyuan75128,000
#102Kimi K2Moonshot AI / Kimi75131,072
#103Kimi K2 ThinkingMoonshot AI / Kimi75256,000
#104Kimi K2 Turbo PreviewMoonshot AI / Kimi75256,000
#105Kimi K2.5Moonshot AI / Kimi75256,000
#106Magistral Medium 1.2Mistral AI75128,000
#107MiniMax-M1MiniMax75204,800
#108MiniMax-M2MiniMax75204,800
#109MiniMax-M2.1MiniMax75204,800
#110MiniMax-M2.1-highspeedMiniMax75204,800
#111MiniMax-M2.5MiniMax75204,800
#112MiniMax-M2.5-highspeedMiniMax75204,800
#113MiniMax-Text-01MiniMax75204,800
#114MiniMax-VL-01MiniMax75204,800
#115Mistral EmbedMistral AI7532,768
#116Mistral Large 3Mistral AI75128,000
#117Mistral Medium 3.1Mistral AI75128,000
#118Mistral ModerationMistral AI7532,768
#119Mistral Small 3.1Mistral AI75128,000
#120Mistral Small 3.2 OpenMistral AI75128,000
#121Pixtral 12BMistral AI75131,072
#122Pixtral LargeMistral AI75131,072
#123Vidu Q1Z.AI75128,000
#124Voxtral Mini OpenMistral AI75131,072
#125Voxtral Small OpenMistral AI75131,072
#126Claude Sonnet 3Anthropic74200,000
#127Claude Sonnet 4Anthropic741,000,000
#128Command ACohere74256,000
#129Command A ReasoningCohere74128,000
#130Command A TranslateCohere74128,000
#131Command A VisionCohere74128,000
#132Command R+Cohere74128,000
#133Command R7BCohere74128,000
#134DeepSeek-Coder-V2DeepSeek74128,000
#135DeepSeek-Math-V2DeepSeek74128,000
#136DeepSeek-R1DeepSeek74128,000
#137DeepSeek-R1-Distill-Llama-70BDeepSeek74128,000
#138DeepSeek-V2.5DeepSeek74128,000
#139DeepSeek-V3DeepSeek74128,000
#140DeepSeek-V3.1DeepSeek74128,000
#141DeepSeek-V3.1-BaseDeepSeek74128,000
#142DeepSeek-V3.2DeepSeek74128,000
#143DeepSeek-V3.2-ExpDeepSeek74128,000
#144Devstral 2 OpenMistral AI74128,000
#145Devstral Small 2Mistral AI74128,000
#146Embed 4Cohere74128,000
#147ERNIE 4.5 Turbo 32KBaidu / ERNIE7432,768
#148Grok 3xAI74131,072
#149Grok 3 MinixAI74131,072
#150Grok 4xAI74256,000
#151Grok 4 Fast ReasoningxAI74131,072
#152grok-imagexAI74131,072
#153image-01MiniMax748,192
#154image-01-liveMiniMax748,192
#155Jamba 3BAI21 Labs74256,000
#156Jamba LargeAI21 Labs74256,000
#157Jamba Large 1.6AI21 Labs74256,000
#158Jamba MiniAI21 Labs74256,000
#159Jamba Mini 1.6AI21 Labs74256,000
#160Jamba Mini 1.7AI21 Labs74256,000
#161Magistral Small 1.2 OpenMistral AI74128,000
#162MiniMax-Speech-02MiniMax748,192
#163Ministral 3 14B OpenMistral AI74128,000
#164Ministral 3 3B OpenMistral AI74128,000
#165Ministral 3 8B OpenMistral AI74128,000
#166Mistral Large 3 OpenMistral AI74128,000
#167Mistral Nemo 12BMistral AI74128,000
#168Mistral OCR 2505Mistral AI7432,768
#169music-2.0MiniMax748,192
#170Phi-3-vision-128k-instructMicrosoft74128,000
#171Phi-3.5-mini-instructMicrosoft74131,072
#172Phi-3.5-MoE-instructMicrosoft74131,072
#173Phi-3.5-vision-instructMicrosoft74131,072
#174Phi-4-mini-flash-reasoningMicrosoft74131,072
#175Phi-4-mini-instructMicrosoft74131,072
#176Phi-4-multimodal-instructMicrosoft74131,072
#177Phi-4-reasoningMicrosoft74131,072
#178Phi-4-reasoning-plusMicrosoft74131,072
#179Phi-4-reasoning-vision-15BMicrosoft74131,072
#180Qwen2.5-1.5B-InstructAlibaba Qwen74131,072
#181Qwen2.5-14B-InstructAlibaba Qwen74131,072
#182Qwen2.5-32B-InstructAlibaba Qwen74131,072
#183Qwen2.5-3B-InstructAlibaba Qwen74131,072
#184Qwen2.5-72B-InstructAlibaba Qwen74131,072
#185Qwen2.5-7B-InstructAlibaba Qwen74131,072
#186Qwen2.5-MaxAlibaba Qwen74131,072
#187Qwen2.5-VL-72B-InstructAlibaba Qwen74131,072
#188Qwen2.5-VL-7B-InstructAlibaba Qwen74131,072
#189Qwen3-Coder-NextAlibaba Qwen74131,072
#190Qwen3.5-0.8BAlibaba Qwen74131,072
#191Qwen3.5-122B-A10BAlibaba Qwen74131,072
#192Qwen3.5-27BAlibaba Qwen74131,072
#193Qwen3.5-2BAlibaba Qwen74131,072
#194Qwen3.5-35B-A3BAlibaba Qwen74131,072
#195Qwen3.5-397B-A17BAlibaba Qwen74131,072
#196Qwen3.5-4BAlibaba Qwen74131,072
#197Qwen3.5-9BAlibaba Qwen74131,072
#198DeepSeek-OCRDeepSeek7316,384
#199DeepSeek-OCR-2DeepSeek7316,384
#200DeepSeek-VL2-SmallDeepSeek7316,384
#201GPT-4o mini TranscribeOpenAI73128,000
#202GPT-4o mini TTSOpenAI73128,000
#203GPT-4o TranscribeOpenAI73128,000
#204gpt-oss-120bOpenAI73131,072
#205Janus-Pro-7BDeepSeek7316,384
#206Phi-4Microsoft7316,384
#207Claude Opus 3Anthropic72200,000
#208Claude Opus 4Anthropic72200,000
#209Llama 3.3 70B InstructMeta72128,000
#210phi-1Microsoft724,096
#211phi-1_5Microsoft724,096
#212phi-2Microsoft724,096
#213Phi-3-medium-4k-instructMicrosoft724,096
#214Phi-3-mini-4k-instructMicrosoft724,096
#215Phi-tiny-MoE-instructMicrosoft724,096
#216pplx-embed-v1-4bPerplexity728,192
#217Prompt Guard 86MMeta72512
#218pplx-embed-v1-0.6bPerplexity718,192
#219Step-3.5-FlashStepFun71131,072
#220gpt-oss-20bOpenAI70131,072
#221Llama 3.1 405B InstructMeta70128,000
#222Llama 4 ScoutMeta7010,485,760
#223LFM2-24B-A2BLiquid AI6932,768
#224Llama 3.1 70B InstructMeta69128,000
#225MiMo-VL-7BXiaomi68131,072
#226Step3-VL-10BStepFun68131,072
#227Llama 3.1 8B InstructMeta67128,000
#228Llama 3.2 90B Vision InstructMeta67128,000
#229MiMo-Audio-7BXiaomi67131,072
#230LFM2-8B-A1BLiquid AI6632,768
#231Llama 3.2 1B InstructMeta66128,000
#232Llama 3.2 3B InstructMeta66128,000
#233Step-Audio-R1.1StepFun66131,072
#234Llama 3.2 11B Vision InstructMeta65128,000
#235Code Llama 70B InstructMeta648,192
#236Meta Llama 3 70B InstructMeta648,192
#237FLUX 1.1 ProBlack Forest Labs63512
#238LFM2.5-1.2B-InstructLiquid AI63131,072
#239LFM2.5-1.2B-ThinkingLiquid AI63131,072
#240Code Llama 34B InstructMeta628,192
#241FLUX 1.1 Pro UltraBlack Forest Labs62512
#242Meta Llama 3 8B InstructMeta628,192
#243FLUX 1 ProBlack Forest Labs61512
#244LFM2-2.6BLiquid AI6032,768
#245NextStep-1.1StepFun54512

Why #1: Claude Opus 4.6

Anthropic's most intelligent Claude model for complex agents, coding, and deep reasoning, with 1M token context and 128K output.

This model clears the current full-profile threshold for leaderboard methodology.

Why #2: Claude Sonnet 4.5

Anthropic's Sonnet 4.5 with 1M token context for fast frontier reasoning, coding, and long-context agent work.

This model clears the current full-profile threshold for leaderboard methodology.

Why #3: Claude Sonnet 4.6

Anthropic's current Sonnet tier for fast frontier reasoning, coding, and long-context agent work.

This model clears the current full-profile threshold for leaderboard methodology.