AI Model Ranking (LLM Leaderboard)
Cheapest AI Models
Most affordable language models sorted by price per token
|
Model
AI model name and provider organization |
Input/1M
Cost per 1 million input tokens (text you send to the model) |
Output/1M
Cost per 1 million output tokens (text the model generates for you) |
MMLU-Pro
Massive Multitask Language Understanding (Professional) - tests broad knowledge across 14 subjects including STEM, humanities, and social sciences |
GPQA
Graduate-level Google-Proof Q&A benchmark - tests PhD-level reasoning and advanced intelligence |
AIME 2025
American Invitational Mathematics Examination 2025 - tests advanced mathematical problem-solving ability |
Release
When the model was released - newer models may have more capabilities | Compare |
|---|---|---|---|---|---|---|---|
 #1 Grok-1 by xAI | N/A | N/A | - | - | - | Mar 17, 2024 | |
 #2 Gemma 3 27B Instruct by Google | N/A | N/A | 66.9% | 42.8% | 20.7% | Mar 12, 2025 | |
| #3 Gemma 3 270M by Google | N/A | N/A | 5.5% | 22.4% | 2.3% | Aug 14, 2025 | |
| #4 Gemma 3n E2B Instruct by Google | N/A | N/A | 37.8% | 22.9% | 10.3% | Jun 26, 2025 | |
| #5 Gemma 3 12B Instruct by Google | N/A | N/A | 59.5% | 34.9% | 18.3% | Mar 12, 2025 | |
| #6 Gemma 3 4B Instruct by Google | N/A | N/A | 41.7% | 29.1% | 12.7% | Mar 12, 2025 | |
| #7 Gemma 3 1B Instruct by Google | N/A | N/A | 13.5% | 23.7% | 3.3% | Mar 13, 2025 | |
 #8 Devstral 2 by Mistral | N/A | N/A | 76.2% | 59.4% | 36.7% | Dec 9, 2025 | |
| #9 Devstral Small 2 by Mistral | N/A | N/A | 67.8% | 53.2% | 34.3% | Dec 9, 2025 | |
 #10 DeepSeek V3.2 Speciale by DeepSeek | N/A | N/A | 86.3% | 87.1% | 96.7% | Dec 1, 2025 | |
| #11 DeepSeek R1 0528 Qwen3 8B by DeepSeek | N/A | N/A | 73.9% | 61.2% | 63.7% | May 29, 2025 | |
 #12 R1 1776 by Perplexity | N/A | N/A | - | - | - | Feb 18, 2025 | |
 #13 Falcon-H1R-7B by TII UAE | N/A | N/A | 72.5% | 66.1% | 80.0% | Jan 4, 2026 | |
| #14 Grok Voice Agent by xAI | N/A | N/A | - | - | - | Dec 17, 2025 | |
 #15 Phi-4 Mini Instruct by Microsoft Azure | N/A | N/A | 46.5% | 33.1% | 6.7% | Feb 26, 2024 | |
| #16 Phi-4 Multimodal Instruct by Microsoft Azure | N/A | N/A | 48.5% | 31.5% | - | Feb 26, 2025 | |
 #17 LFM2.5-VL-1.6B by Liquid AI | N/A | N/A | - | 28.9% | - | Jan 5, 2026 | |
| #18 LFM2.5-1.2B-Thinking by Liquid AI | N/A | N/A | - | 33.9% | - | Jan 20, 2026 | |
| #19 LFM2 8B A1B by Liquid AI | N/A | N/A | 50.5% | 34.4% | 25.3% | Oct 7, 2025 | |
| #20 LFM2 2.6B by Liquid AI | N/A | N/A | 29.8% | 30.6% | 8.3% | Sep 23, 2025 | |
| #21 LFM2.5-1.2B-Instruct by Liquid AI | N/A | N/A | - | 32.6% | - | Jan 5, 2026 | |
 #22 Solar Pro 2 (Non-reasoning) by Upstage | N/A | N/A | 75.0% | 56.1% | 30.0% | Jul 9, 2025 | |
| #23 Solar Pro 2 (Reasoning) by Upstage | N/A | N/A | 80.5% | 68.7% | 61.3% | Jul 9, 2025 | |
| #24 Solar Open 100B (Reasoning) by Upstage | N/A | N/A | - | 65.7% | - | Dec 17, 2025 | |
 #25 Llama 3.3 Nemotron Super 49B v1 (Reasoning) by NVIDIA | N/A | N/A | 78.5% | 64.3% | 54.7% | Mar 18, 2025 | |
| #26 Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning) by NVIDIA | N/A | N/A | 55.6% | 40.8% | 50.0% | May 20, 2025 | |
| #27 Llama 3.3 Nemotron Super 49B v1 (Non-reasoning) by NVIDIA | N/A | N/A | 69.8% | 51.7% | 7.7% | Mar 18, 2025 | |
 #28 Kimi Linear 48B A3B Instruct by Kimi | N/A | N/A | 58.5% | 41.2% | 36.3% | Oct 30, 2025 | |
 #29 Step3 VL 10B by StepFun | N/A | N/A | - | 69.0% | - | Jan 20, 2026 | |
 #30 Molmo 7B-D by Allen Institute for AI | N/A | N/A | 37.1% | 24.0% | - | Sep 25, 2024 | |
| #31 Molmo2-8B by Allen Institute for AI | N/A | N/A | - | 42.5% | - | Dec 11, 2025 | |
| #32 Olmo 3.1 32B Think by Allen Institute for AI | N/A | N/A | 76.3% | 59.1% | 77.3% | Dec 12, 2025 | |
 #33 Granite 4.0 1B by IBM | N/A | N/A | 32.5% | 28.1% | 6.3% | Oct 28, 2025 | |
| #34 Granite 4.0 Micro by IBM | N/A | N/A | 44.7% | 33.6% | 6.0% | Sep 22, 2025 | |
| #35 Granite 4.0 350M by IBM | N/A | N/A | 12.4% | 26.1% | - | Oct 28, 2025 | |
| #36 Granite 4.0 H 350M by IBM | N/A | N/A | 12.7% | 25.7% | 1.3% | Oct 28, 2025 | |
| #37 Granite 4.0 H 1B by IBM | N/A | N/A | 27.7% | 26.3% | 6.3% | Oct 28, 2025 | |
 #38 DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning) by Nous Research | N/A | N/A | 36.5% | 27.0% | - | Feb 13, 2025 | |
| #39 DeepHermes 3 - Mistral 24B Preview (Non-reasoning) by Nous Research | N/A | N/A | 58.0% | 38.2% | - | Mar 13, 2025 | |
 #40 K-EXAONE (Reasoning) by LG AI Research | N/A | N/A | 83.8% | 78.3% | 90.3% | Dec 31, 2025 | |
| #41 K-EXAONE (Non-reasoning) by LG AI Research | N/A | N/A | 81.0% | 69.5% | 44.0% | Dec 31, 2025 | |
| #42 Exaone 4.0 1.2B (Non-reasoning) by LG AI Research | N/A | N/A | 50.0% | 42.4% | 24.0% | Jul 15, 2025 | |
| #43 Exaone 4.0 1.2B (Reasoning) by LG AI Research | N/A | N/A | 58.8% | 51.5% | 50.3% | Jul 15, 2025 | |
 #44 ERNIE 5.0 Thinking Preview by Baidu | N/A | N/A | 83.0% | 77.7% | 85.0% | Nov 13, 2025 | |
 #45 Llama 65B by Meta | N/A | N/A | - | - | - | Feb 24, 2023 | |
 #46 INTELLECT-3 by Prime Intellect | N/A | N/A | 82.2% | 76.1% | 88.0% | Nov 27, 2025 | |
 #47 Motif-2-12.7B-Reasoning by Motif Technologies | N/A | N/A | 79.6% | 69.5% | 80.3% | Dec 4, 2025 | |
 #48 K2-V2 (medium) by MBZUAI Institute of Foundation Models | N/A | N/A | 76.1% | 59.8% | 64.7% | Dec 5, 2025 | |
| #49 K2-V2 (low) by MBZUAI Institute of Foundation Models | N/A | N/A | 71.3% | 54.1% | 35.3% | Dec 5, 2025 | |
| #50 K2-V2 (high) by MBZUAI Institute of Foundation Models | N/A | N/A | 78.6% | 68.1% | 78.3% | Dec 5, 2025 | |
| #51 K2 Think V2 by MBZUAI Institute of Foundation Models | N/A | N/A | - | 71.3% | - | Dec 15, 2025 | |
 #52 Mi:dm K 2.5 Pro by Korea Telecom | N/A | N/A | 80.9% | 70.1% | 76.7% | Dec 11, 2025 | |
| #53 Mi:dm K 2.5 Pro Preview by Korea Telecom | N/A | N/A | 81.3% | 72.2% | 78.7% | Dec 11, 2025 | |
 #54 HyperCLOVA X SEED Think (32B) by Naver | N/A | N/A | 78.5% | 61.5% | 59.0% | Dec 26, 2025 | |
 #55 Tri-21B-Think by Trillion Labs | N/A | N/A | - | 60.1% | - | Feb 10, 2026 | |
| #56 Tri-21B-think Preview by Trillion Labs | N/A | N/A | - | 53.8% | - | Feb 10, 2026 | |
 #57 Tiny Aya Global by Cohere | N/A | N/A | - | 30.5% | - | Feb 17, 2026 | |
 #58 Apriel-v1.6-15B-Thinker by ServiceNow | N/A | N/A | 79.0% | 73.3% | 88.0% | Nov 25, 2025 | |
 #59 Jamba Reasoning 3B by AI21 Labs | N/A | N/A | 57.7% | 33.3% | 10.7% | Oct 8, 2025 | |
| #60 Jamba 1.7 Mini by AI21 Labs | N/A | N/A | 38.8% | 32.2% | 0.3% | Jul 7, 2025 | |
 #61 Qwen Chat 14B by Alibaba | N/A | N/A | - | - | - | Sep 25, 2023 | |
| #62 Qwen3 4B 2507 (Reasoning) by Alibaba | N/A | N/A | 74.3% | 66.7% | 82.7% | Aug 6, 2025 | |
| #63 Qwen3 VL 4B Instruct by Alibaba | N/A | N/A | 63.4% | 37.1% | 37.0% | Oct 14, 2025 | |
| #64 Qwen3 VL 4B (Reasoning) by Alibaba | N/A | N/A | 70.0% | 49.4% | 25.7% | Oct 14, 2025 | |
| #65 Qwen3 4B 2507 Instruct by Alibaba | N/A | N/A | 67.2% | 51.7% | 52.3% | Aug 6, 2025 | |
 #66 Ring-1T by InclusionAI | N/A | N/A | 80.6% | 77.4% | 89.3% | Oct 13, 2025 | |
| #67 Ling-1T by InclusionAI | N/A | N/A | 82.2% | 71.9% | 71.3% | Oct 8, 2025 | |
 #68 Doubao Seed 2.0 lite (Reasoning) by ByteDance Seed | N/A | N/A | - | 65.6% | - | Feb 15, 2026 | |
| #69 Doubao Seed Code by ByteDance Seed | N/A | N/A | 85.4% | 76.4% | 79.3% | Nov 11, 2025 | |
 #70 o1-mini by OpenAI | N/A | N/A | 74.2% | 60.3% | - | Sep 12, 2024 | |
| #71 GPT-4o (ChatGPT) by OpenAI | N/A | N/A | 77.3% | 51.1% | - | Feb 15, 2025 | |
| #72 GPT-4o mini Realtime (Dec '24) by OpenAI | N/A | N/A | - | - | - | Dec 17, 2024 | |
| #73 GPT-4o (March 2025, chatgpt-4o-latest) by OpenAI | N/A | N/A | 80.3% | 65.5% | 25.7% | Mar 27, 2025 | |
| #74 GPT-3.5 Turbo (0613) by OpenAI | N/A | N/A | - | - | - | Jun 13, 2023 | |
| #75 GPT-4o Realtime (Dec '24) by OpenAI | N/A | N/A | - | - | - | Dec 17, 2024 | |
| #76 GPT-4.5 (Preview) by OpenAI | N/A | N/A | - | - | - | Feb 27, 2025 | |
| #77 Llama 2 Chat 70B by Meta | N/A | N/A | 40.6% | 32.7% | - | Jul 18, 2023 | |
| #78 Llama 2 Chat 13B by Meta | N/A | N/A | 40.6% | 32.1% | - | Jul 18, 2023 | |
| #79 Gemini 2.0 Pro Experimental (Feb '25) by Google | N/A | N/A | 80.5% | 62.2% | - | Feb 5, 2025 | |
| #80 Gemini 2.0 Flash (experimental) by Google | N/A | N/A | 78.2% | 63.6% | - | Dec 11, 2024 | |
| #81 Gemini 1.5 Pro (Sep '24) by Google | N/A | N/A | 75.0% | 58.9% | - | Sep 24, 2024 | |
| #82 Gemini 2.0 Flash-Lite (Preview) by Google | N/A | N/A | - | 54.2% | - | Feb 5, 2025 | |
| #83 Gemini 1.5 Flash (Sep '24) by Google | N/A | N/A | 68.0% | 46.3% | - | Sep 24, 2024 | |
| #84 Gemini 1.5 Flash-8B by Google | N/A | N/A | 56.9% | 35.9% | - | Oct 3, 2024 | |
| #85 PALM-2 by Google | N/A | N/A | - | - | - | May 10, 2023 | |
| #86 Gemini 2.0 Flash-Lite (Feb '25) by Google | N/A | N/A | 72.4% | 53.5% | - | Feb 25, 2025 | |
| #87 Gemini 1.0 Ultra by Google | N/A | N/A | - | - | - | Dec 6, 2023 | |
| #88 Gemini 1.0 Pro by Google | N/A | N/A | 43.1% | 27.7% | - | Dec 6, 2023 | |
| #89 Gemini 1.5 Flash (May '24) by Google | N/A | N/A | 57.4% | 32.4% | - | May 14, 2024 | |
| #90 Gemma 3n E4B Instruct Preview (May '25) by Google | N/A | N/A | 48.3% | 27.8% | - | May 20, 2025 | |
| #91 Gemini 2.0 Flash Thinking Experimental (Dec '24) by Google | N/A | N/A | - | - | - | Dec 19, 2024 | |
| #92 Gemini 2.0 Flash Thinking Experimental (Jan '25) by Google | N/A | N/A | 79.8% | 70.1% | - | Jan 21, 2025 | |
| #93 Gemini 2.5 Flash Preview (Non-reasoning) by Google | N/A | N/A | 78.3% | 59.4% | - | Apr 17, 2025 | |
| #94 Gemini 2.5 Flash Preview (Reasoning) by Google | N/A | N/A | 80.0% | 69.8% | - | Apr 17, 2025 | |
| #95 Gemini 2.5 Pro Preview (Mar' 25) by Google | N/A | N/A | 85.8% | 83.6% | - | Mar 25, 2025 | |
| #96 Gemini 1.5 Pro (May '24) by Google | N/A | N/A | 65.7% | 37.1% | - | May 15, 2024 | |
 #97 Claude Instant by Anthropic | N/A | N/A | 43.4% | 33.0% | - | Mar 14, 2023 | |
| #98 Claude 2.1 by Anthropic | N/A | N/A | 49.5% | 31.9% | - | Nov 21, 2023 | |
| #99 Claude 2.0 by Anthropic | N/A | N/A | 48.6% | 34.4% | - | Jul 11, 2023 | |
| #100 Mixtral 8x22B Instruct by Mistral | N/A | N/A | 53.7% | 33.2% | - | Apr 17, 2024 | |
Understanding the AI Model Leaderboard
This comprehensive AI model leaderboard helps you compare and choose the best large language models (LLMs) for your needs. We track standardized AI benchmarks, token pricing, inference speed, and model capabilities across all major AI providers like OpenAI, Anthropic, Google, Meta, and DeepSeek.
Core AI Benchmarks Explained
- MMLU-Pro: Tests broad knowledge across 14 academic subjects including STEM, humanities, and social sciences - the foundational intelligence benchmark
- GPQA: Graduate-level Google-Proof Q&A benchmark - measures PhD-level reasoning and advanced problem-solving capabilities
- AIME 2025: American Invitational Mathematics Examination - evaluates elite mathematical reasoning and competition-level problem solving
- Coding Index: Composite score of LiveCodeBench, SciCode, and coding benchmarks - measures programming ability
- Math Index: Composite score of AIME, MATH-500, and mathematical reasoning tests
Key Metrics to Consider
- Token Pricing: Compare input vs output token costs per million - crucial for estimating API expenses and optimizing usage patterns
- Inference Speed: Measured in tokens/second - determines response time for chatbots, streaming, and real-time applications
- Release Date: Newer models often incorporate latest training techniques and updated knowledge cutoffs
- Benchmark Scores: Percentage scores (0-100%) make it easy to compare model capabilities at a glance
How to Choose the Right AI Model for Your Use Case
For Research & Analysis
Prioritize models with high MMLU-Pro (70%+) and GPQA (60%+) scores for complex reasoning tasks, academic research, and technical documentation
For Cost Optimization
Sort by input/output pricing - smaller models often deliver 80% of flagship performance at 10% of the cost for simple tasks
For Math & STEM
Filter by Math Index or AIME 2025 scores (50%+) for quantitative analysis, engineering calculations, and scientific applications
All benchmark scores and pricing data are updated daily from Artificial Analysis to reflect the latest model versions and capabilities. Use the sort filters above to find AI models by intelligence, cost, coding ability, math performance, speed, or release date.
Frequently Asked Questions
What is MMLU-Pro and why is it the standard AI intelligence benchmark?
MMLU-Pro (Massive Multitask Language Understanding - Professional) is the most comprehensive AI benchmark, testing models across 14 academic subjects including mathematics, science, history, law, and ethics. Scores range from 46% (basic competency) to 87% (near-expert level). Models scoring above 75% demonstrate strong general intelligence suitable for professional applications, while scores below 60% indicate limitations in complex reasoning tasks.
What does GPQA measure and which models score highest?
GPQA (Graduate-level Google-Proof Q&A) tests PhD-level reasoning with questions designed to be "Google-proof" - requiring deep understanding rather than simple fact retrieval. Top models like GPT-5.1 (87.3%), GPT-5 mini (82.8%), and o3 (82.7%) excel at GPQA, making them ideal for research, technical analysis, and complex problem-solving. Models below 50% GPQA struggle with advanced reasoning and may provide superficial answers to complex questions.
What is AIME 2025 and how does it evaluate AI mathematical ability?
AIME 2025 (American Invitational Mathematics Examination) is an elite math competition benchmark that tests advanced problem-solving, algebra, geometry, and number theory. Scores above 80% (like GPT-5 Codex at 98.7% or GPT-5.1 at 94%) indicate exceptional mathematical reasoning suitable for engineering, scientific computing, and quantitative analysis. Models scoring below 50% may struggle with multi-step mathematical problems or require explicit problem breakdown.
How is AI model pricing calculated and what's considered cost-effective?
AI model pricing is measured per 1 million tokens (approximately 750,000 words). Input pricing covers text you send, while output pricing covers generated responses. Budget models like Llama 3.3 70B cost $0.54/$0.71 per million tokens, mid-tier models like GPT-5 nano cost $0.05/$0.40, while premium models like GPT-5 cost $1.25/$10. For typical applications with 3:1 input-to-output ratio, budget models can be 10-20x cheaper than flagship models while maintaining 70-80% performance.
Which AI models are best for coding and programming tasks?
Sort by Coding Index to see top programming models. Our Coding Index combines LiveCodeBench, SciCode, and coding benchmarks. Top performers include GPT-5.1 (57.5 index), GPT-5 mini (51.4), and GPT-5 Codex (53.5). These models excel at code generation, debugging, refactoring, and explaining complex algorithms. For budget-conscious developers, models with 40+ coding index scores offer excellent value for routine programming tasks.
How often are AI model benchmarks and rankings updated?
Our leaderboard syncs daily with Artificial Analysis API to ensure benchmark scores (MMLU-Pro, GPQA, AIME 2025), pricing, and inference speed data reflect the latest model versions. New model releases appear immediately under the "Newest" sort option. Benchmark scores can change when providers release updated versions - for example, GPT-5.1 released in November 2025 achieved 69.7 intelligence compared to GPT-5's 68.5 from August 2025.
What inference speed (tokens/second) do I need for my application?
Inference speed determines how fast models generate responses. For real-time chatbots and interactive applications, target 100+ tokens/second (models like gpt-oss-120B at 340 tok/s). For background processing and batch jobs, 50-100 tok/s is sufficient. Premium reasoning models like GPT-5 (103 tok/s) balance speed and capability. Note that higher inference speed doesn't always mean better quality - slower models often deliver more thoughtful, detailed responses.
Can I test these AI models for free before committing?
Yes! Try our free AI chat interface to test different models instantly without creating an account. Many providers also offer free tiers: OpenAI (ChatGPT with daily limits), Anthropic (Claude with usage caps), Google (Gemini free tier), and open-source models like Llama 3.3. Compare performance on your specific use case before upgrading to paid plans.