Leaderboard

    Best AI for Science 2026.

    Find the best AI for science and research. Ranked by GPQA Diamond, FrontierMath, Frontier Science Research, CharXiv, and MMMU-Pro benchmarks.

    LLMImageVideoSpeechTranscription
    Gemini 3 Pro Preview
    Google
    		91.9%91.9%81%81.4%———
    gpt-5.4-pro
    OpenAI
    		—90.5%——17.4%—10.8%
    gpt-5.4
    OpenAI
    		92.8%89.4%81.2%—15.6%47.6%9.4%
    gpt-5.1
    OpenAI
    		88.1%88.1%———26.7%—
    gpt-5.2-pro
    OpenAI
    		93.2%87.9%——14.9%—8.7%
    gpt-5.3-codex
    OpenAI
    		—87.7%——14.6%—7.3%
    Grok 4
    xAI
    		87.5%87.5%—————
    gpt-5
    OpenAI
    		85.7%87.3%78.4%81.1%—26.3%—
    Claude Opus 4.5
    Anthropic
    		87%87%—————
    Gemini 2.5 Pro
    Google
    		83%86.4%—————
    gpt-5.2
    OpenAI
    		92.4%85.4%79.5%82.1%12.9%40.3%7.6%
    Grok 3
    xAI
    		84.6%84.6%—————
    Kimi K2 Thinking
    Moonshot AI
    		—84.5%—————
    Claude Sonnet 4.5
    Anthropic
    		83.4%83.4%—————
    o3
    OpenAI
    		83.3%83.3%76.4%78.6%10.3%15.8%0%
    o4-mini
    OpenAI
    		81.4%81.4%—72%———
    o3-mini
    OpenAI
    		77.2%79.7%———9.2%—
    Claude Opus 4
    Anthropic
    		79.6%79.6%—————
    Gemini 2.5 Flash
    Google
    		82.8%78.3%—————
    o1
    OpenAI
    		78%75.7%———5.5%—
    Claude Sonnet 4
    Anthropic
    		75.4%75.4%—————
    Claude Haiku 4.5
    Anthropic
    		73%73%—————
    DeepSeek R1OSS
    DeepSeek
    		—71.5%—————
    gpt-4.5-preview
    OpenAI
    		—71.4%—————
    Llama 4 MaverickOSS
    Meta
    		69.8%69.8%59.6%————
    Claude Sonnet 3.7
    Anthropic
    		84.8%68%—————
    gpt-4.1-mini
    OpenAI
    		65%65%—56.8%———
    Claude 3.5 Sonnet
    Anthropic
    		67.2%65%—————
    DeepSeek V3OSS
    DeepSeek
    		59.1%64.8%—————
    Gemini 2.0 Flash
    Google
    		62.1%62.1%—————
    o1 mini
    OpenAI
    		60%60%—————
    Llama 4 ScoutOSS
    Meta
    		57.2%57.2%—————
    GPT-4o
    OpenAI
    		70.1%56.1%59.9%58.8%———
    Llama 3.3 70B
    Meta
    		—50.5%—————
    gpt-4.1-nano
    OpenAI
    		50.3%50.3%—40.5%———
    Llama 3.1 405B
    Meta
    		—49%—————
    Amazon Nova Pro
    Amazon
    		—46.9%—————
    gpt-4.1
    OpenAI
    		66.3%43.4%—56.7%13.2%——
    Claude 3.5 Haiku
    Anthropic
    		41.6%41.6%—————
    GPT-4o-mini
    OpenAI
    		40.2%40.2%—————
    Amazon Nova 2 Lite
    Amazon
    		———————
    Amazon Nova 2 Omni
    Amazon
    		———————
    Amazon Nova 2 Pro
    Amazon
    		———————
    Amazon Nova Lite
    Amazon
    		———————
    Amazon Nova Micro
    Amazon
    		———————
    Amazon Nova Premier
    Amazon
    		———————
    Claude 2
    Anthropic
    		———————
    Claude 2.1
    Anthropic
    		———————
    Claude 3 Haiku
    Anthropic
    		33.3%——————
    Claude 3 Opus
    Anthropic
    		50.4%——————
    Showing 1–50 of 336 models

    Building with these APIs?

    Get 10+ Next.js AI templates with auth, payments, and more.

    Get Templates — $249

    All Large Language Models

    OpenAI

    60 models
    gpt-5.4·gpt-5.4-pro·gpt-5.3-chat-latest·gpt-5.3-instant·gpt-5.3-codex·GPT-5.2 Codex·gpt-5.2·gpt-5.1-codex·gpt-5.1-codex-mini·GPT-5.1 Codex High·GPT-5.1 High·GPT-5.1 Instant·GPT-5.1 Medium·GPT-5.1 Thinking·gpt-5-codex·gpt-5.2-pro·gpt-5·GPT-5 High·GPT-5 Medium·gpt-5-mini·gpt-5-nano·GPT OSS 120B·GPT OSS 120B High·GPT OSS 20B·GPT OSS 20B High·o3-pro·o3·o4-mini·gpt-4.1·gpt-4.1-mini·gpt-4.1-nano·o1-pro·gpt-4.5-preview·o3-mini·o1·gpt-5-pro·gpt-5.1·o1 mini·o1 preview·o1-mini-2024-09-12·o1-preview-2024-09-12·GPT-4o-2024-08-06·GPT-4o-mini·GPT-4o·GPT-4o-2024-05-13·GPT-4-turbo-2024-04-09·GPT-3.5-turbo-0125·GPT-4-turbo-0125·GPT-3.5-turbo-1106·GPT-4 Turbo·GPT-4-turbo-1106·computer-use-preview·gpt-4o-audio-preview·GPT-3.5-turbo-16k·GPT-4-0613·GPT-4-32k-0613·GPT-3.5 Turbo·GPT-4·GPT-4-32k·GPT-3.5-turbo

    Anthropic

    21 models
    Claude Sonnet 4.6·Claude Opus 4.6·Claude Opus 4.5·Claude Haiku 4.5·Claude Sonnet 4.5·Claude Opus 4.1·Claude Opus 4·Claude Sonnet 4·Claude Sonnet 3.7·Claude Haiku 3.5·Claude 3.5 Haiku·Claude 3.5 Sonnet·Claude 3 Haiku·Claude Haiku 3·Claude 3 Opus·Claude 3 Sonnet·Claude Opus 3·Claude 2.1·Claude Instant 1.2·Claude 2·Claude Sonnet 3.5

    Google

    20 models
    Gemini 3.1 Flash Lite Preview·Gemini 3.1 Pro·Gemini 3 Flash Preview·Gemini 3 Pro Preview·Gemma 3n E4B Instructed·Gemini 2.5 Flash Lite·Gemini 2.5 Pro·Gemini 2.5 Pro Preview 06-05·Gemini 2.5 Flash·Gemma 3 12B·Gemma 3 27B·Gemma 3 4B·Gemini 2.0 Flash·Gemini 2.0 Flash Lite·Gemini 1.5 Pro·Gemini 1.5 Flash·Gemini 1.5 Flash·Gemini 1.5 Flash 8B·Gemini 1.0 Pro·Gemini 1.5 Pro

    Perplexity

    5 models
    Llama 3.1 70b·Sonar·R1 1776·Sonar Deep Research·Sonar Pro

    xAI

    16 models
    Grok-4.1·Grok-4.1 Fast Non-Reasoning·Grok-4.1 Fast Reasoning·Grok-4.1 Thinking·Grok 4 Fast·Grok Code Fast 1·Grok-4 Fast Non-Reasoning·Grok-4 Fast Reasoning·Grok 4·Grok 3·Grok 3 Mini·Grok 2·Grok 2 Vision·Grok 3 Fast·Grok 3 Mini Fast·Grok 4.1 Fast

    DeepSeek

    13 models
    deepseek-chat·DeepSeek-V3.2-Exp·DeepSeek-R1-0528·DeepSeek-V3 0324·DeepSeek R1·DeepSeek R1 Distill Llama 70B·DeepSeek R1 Distill Qwen 32B·DeepSeek V3.1·DeepSeek V3·DeepSeek-V2.5·DeepSeek V2·DeepSeek V3.2·deepseek-coder

    Groq

    9 models
    Llama 4 Maverick (Groq)·Llama 4 Scout (Groq)·DeepSeek R1 (Groq)·Llama 3.3 70B (Groq)·Llama 3.1 70b·Llama 3.1 8B (Groq)·Mixtral 8x7B·Mistral Small 2503 (Groq)·Qwen QwQ 32B (Groq)

    MiniMax

    4 models
    MiniMax M2.5·MiniMax M2.1·MiniMax M2·MiniMax M1 80K

    Qwen

    29 models
    Qwen 3.5 397B A17B·Qwen3 Max·Qwen3 VL 235B A22B Instruct·Qwen3 VL 235B A22B Thinking·Qwen3 VL 30B A3B Instruct·Qwen3 VL 30B A3B Thinking·Qwen3 VL 4B Instruct·Qwen3 VL 4B Thinking·Qwen3 VL 8B Instruct·Qwen3 VL 8B Thinking·Qwen3-Next-80B-A3B-Instruct·Qwen3-Next-80B-A3B-Thinking·Qwen Plus Latest·Qwen3-235B-A22B-Thinking-2507·Qwen3-235B-A22B-Instruct-2507·Qwen 3 14B·Qwen 3 235B A22B·Qwen 3 30B A3B·Qwen 3 32B·Qwen 3 8B·Qwen3-Coder·QwQ-32B-Preview·Qwen 2.5 72B Instruct·Qwen2.5 7B Instruct·Qwen2.5-Coder 32B Instruct·Qwen 3 4B·Qwen Max Latest·Qwen QwQ Plus·Qwen Turbo Latest

    01.ai

    1 models
    Yi-Large

    AI21 Labs

    2 models
    Jamba 1.5 Large·Jamba 1.5 Mini

    Amazon

    10 models
    Amazon Nova 2 Lite·Amazon Nova Lite·Amazon Nova 2 Omni·Amazon Nova Premier·Amazon Nova Micro·Amazon Nova Pro·Nova Lite·Nova Micro·Nova Pro·Amazon Nova 2 Pro

    Anyscale

    2 models
    Llama 3 70b·Mixtral 8x7B

    Baidu

    1 models
    ERNIE 4.5

    Cerebras

    5 models
    Llama 3.3 70B (Cerebras)·DeepSeek R1 (Cerebras)·DeepSeek V3 (Cerebras)·Llama 3.1 8B (Cerebras)·Qwen 2.5 32B (Cerebras)

    Cohere

    4 models
    Command R7B·Command R·Command R+·Command A

    DeepInfra

    6 models
    Llama 3.3 70B (DeepInfra)·Llama 3.1 405B (DeepInfra)·DeepSeek R1 (DeepInfra)·DeepSeek V3 (DeepInfra)·Mistral Small 2501 (DeepInfra)·Qwen 2.5 72B (DeepInfra)

    Fireworks

    10 models
    DeepSeek V3.2 (Fireworks)·DeepSeek R1 (Fireworks)·Llama 4 Maverick (Fireworks)·Llama 3.1 405b·Llama 3.1 405B (Fireworks)·Llama 3.1 70b·Yi-Large·Mixtral 8x7B·Kimi K2.5 (Fireworks)·Llama 4 Scout (Fireworks)

    Hyperbolic

    4 models
    Llama 3.1 405B (Hyperbolic)·DeepSeek R1 (Hyperbolic)·DeepSeek V3 (Hyperbolic)·Qwen 3 235B A22B (Hyperbolic)

    IBM

    1 models
    Granite 3.3 8B Instruct

    IBM Watsonx

    6 models
    Granite 3.1 8B Instruct·Llama 3.1 405b·Llama 3.1 70b·Mistral Large·Granite 3.1 2B Instruct·Granite 3.2 8B Instruct

    Inception

    1 models
    Mercury 2

    LG AI Research

    1 models
    K-EXAONE-236B-A23B

    Meituan

    4 models
    LongCat-Flash-Lite·LongCat-Flash-Thinking-2601·LongCat-Flash-Thinking·LongCat-Flash-Chat

    Meta

    13 models
    Llama 4 Maverick·Llama 4 Scout·Llama 3.3 70B·Llama 3.3 70B Instruct·Llama 3.2 11B Instruct·Llama 3.2 3B Instruct·Llama 3.2 90B Instruct·Llama 3.1 405B·Llama 3.1 405B Instruct·Llama 3.1 70B·Llama 3.1 70B Instruct·Llama 3.1 8B·Llama 3.1 8B Instruct

    Microsoft

    3 models
    Phi-4-multimodal-instruct·Phi 4·Phi-3.5-mini-instruct

    Mistral

    32 models
    Min istral 3 (3B Reasoning 2512)·Ministral 3 (14B Reasoning 2512)·Ministral 3 (8B Reasoning 2512)·Mistral Large 3 (675B Instruct 2512)·Mistral Large 3·Codestral·Devstral Small 1.1·Devstral Medium·Magistral Medium·Mistral Small 3.1·Mistral Small 3.1 24B Base·Mistral Small 3 24B Instruct·Mistral Large 2411·Pixtral Large 2411·Pixtral Large·Ministral 8B Instruct·Mistral Small·Pixtral-12B·Mistral Large 2·Mistral NeMo Instruct·Mistral Large·Mistral 7B Instruct·Magistral Small·Mistral Codestral·Mistral Devstral·Mistral Devstral 2·Mistral Devstral Small 2·Mistral Medium 3·Mistral Ministral 3B·Mistral Ministral 8B·Mistral Nemo·Mistral Nemo 2407

    Moonshot AI

    5 models
    Kimi K2.5·Kimi K2 0905·Kimi K2-Thinking-0905·Kimi K2 Instruct·Kimi K2 Thinking

    Nebius AI Studio

    5 models
    Qwen 3 235B A22B (Nebius)·DeepSeek V3 (Nebius)·DeepSeek R1 (Nebius)·Llama 3.1 405B (Nebius)·Llama 3.1 70B (Nebius)

    Nous Research

    1 models
    Hermes 3 70B

    NVIDIA

    1 models
    Nemotron 3 Nano (30B A3B)

    Reka

    3 models
    Reka Core·Reka Edge·Reka Flash

    Replicate

    5 models
    DeepSeek R1 (Replicate)·Llama 3.1 405b·Llama 3.1 405B (Replicate)·Llama 3.1 70b·Mixtral 8x7B

    SambaNova Cloud

    5 models
    Llama 3.3 70B (SambaNova)·Qwen 2.5 72B (SambaNova)·Llama 3.1 405B (SambaNova)·DeepSeek R1 (SambaNova)·DeepSeek V3 (SambaNova)

    StepFun

    1 models
    Step-3.5-Flash

    Together.AI

    10 models
    Llama 4 Maverick (Together)·Llama 4 Scout (Together)·DeepSeek R1 (Together)·Llama 3.3 70B (Together)·Llama 3.1 405b·Llama 3.1 405B (Together)·Llama 3.1 70b·Mixtral 8x7B·DeepSeek V3 (Together)·Qwen 2.5 72B (Together)

    Writer

    8 models
    Palmyra X 004·Palmyra Fin 32k·Palmyra Med 32k·Palmyra X 002·Palmyra X 002 32k·Palmyra X 003·Palmyra X 32k·Palmyra X5

    Xiaomi

    1 models
    MiMo-V2-Flash

    Z AI

    8 models
    GLM-5·GLM-5 Code·GLM-4.7-Flash·GLM-4.7·GLM-4.6·GLM-4.5V·GLM-4.5·GLM-4.5 Flash