Awesome AI Models Matrix

Resources

• Explore the curated IDE list: Awesome AI IDEs

🚀 A curated awesome list of top AI models and Large Language Models (LLMs) with comprehensive specifications, benchmarks, pricing, and official resources.

A comprehensive, community-driven resource meticulously curated to help developers, researchers, and organizations navigate the rapidly evolving landscape of artificial intelligence. This matrix provides transparent, up-to-date comparisons of 48+ leading AI models across performance benchmarks, pricing structures, deployment options, and licensing terms.

Whether you're building the next breakthrough application, conducting cutting-edge research, or making strategic technology decisions for your enterprise, this guide empowers you with the critical information needed to choose the perfect AI model for your specific needs.

Last Updated: 🗓️ October 7, 2025 Total Models: 48+ models from 15+ companies Data Sources: OpenRouter Rankings, LLM-Stats.com, Official Documentation

---

📋 Table of Contents

• 🎯 About This Matrix
• 🧠 Understanding LLMs
• 🆕 Latest Updates
• 📊 Model Comparison Tables
  • 🔄 Sort by Latest Update
  • 🏢 Sort by Company
  • 🖥️ Sort by Self-Hosting
  • 💰 Sort by Price
• 🎯 Models by Category
• 💻 Coding Models Deep Dive
• 🔗 Official Resources
• 📈 Performance Benchmarks
• 💰 Cost Analysis
• 🖥️ Self-Hosting Guide
• 🔍 Model Selection Guide
• 🎯 Awesome AI IDEs
• 🤝 Contributing
• 📜 License

---

🎯 About This Matrix

This matrix provides a comprehensive overview of the leading AI models and LLMs available in 2025. Whether you're a 👨‍💻 developer, 🔬 researcher, or 🏢 enterprise decision-maker, this guide helps you understand the capabilities, costs, and trade-offs of each model.

✨ What's Included

• 🆕 Latest Versions - Up-to-date information on model releases
• 📊 Benchmarks - HumanEval, MMLU, SWE-bench, and other metrics
• 💰 Pricing - Transparent cost information per 1M tokens
• 🖥️ Self-Hosting - Open-source availability and licensing
• 🔗 Official Links - Direct access to documentation and APIs
• 🎯 Sortable Tables - Filter by update date, company, pricing, and more

---

🧠 Understanding LLMs

What are Large Language Models?

Large Language Models (LLMs) are AI systems trained on vast amounts of text data to understand and generate human-like language. They power applications ranging from 💬 chatbots and 💻 coding assistants to ✍️ content creation and 📊 data analysis.

🔑 Key Concepts

• Parameters 📏 - The size of the model's neural network (billions or trillions of values)
• Context Window 📖 - How much text the model can process at once (measured in tokens)
• Reasoning Models 🧠 - Models that "think step-by-step" before answering
• Multimodal 🎨 - Models that can process text, images, audio, and video
• MoE (Mixture of Experts) 🎭 - Architecture that activates only relevant parts for efficiency
• RAG (Retrieval Augmented Generation) 🔍 - Enhancing models with external knowledge

🌟 Model Generations (2023-2025)

1. GPT-3.5 Era (2022-2023) 📱 - Foundation of conversational AI
2. GPT-4 Era (2023-2024) 🎨 - Multimodal capabilities and improved reasoning
3. Reasoning Era (2024-2025) 🧠 - Step-by-step thinking models (o1, R1)
4. Unified Era (2025) ⚡ - Hybrid models combining speed and reasoning
5. Open-Source Surge (2025) 🆓 - GPT-OSS, DeepSeek, Qwen competing with proprietary

---

🆕 Latest Updates

🔥 Recent Additions (Last 30 Days)

Date	Model	Company	Highlights	Market Share*
2025-10-03	🏢 IBM Granite 4.0	IBM	ISO 42001 certified, Mamba/Transformer hybrid, open weights	-
2025-09-30	🇨🇳 GLM-4.6	Zhipu AI	355B MoE, real-world coding, open weights, $0.13/$0.39	6.0% tool calls
2025-09-30	🔬 DeepSeek-V3.2-Exp	DeepSeek	Sparse Attention (DSA), efficient long-context, MIT license	11.4% overall
2025-09-29	👑 Claude 4.5 Sonnet	Anthropic	#2 Most Used - 35.5B tokens on OpenRouter	13.6% overall
2025-09-23	🤖 GPT-5 Codex	OpenAI	7+ hour autonomous coding, API and open weights	13.8% overall
2025-09-11	🪶 Phi-4	Microsoft	14B compact reasoning, MIT license, self-hostable	-
2025-09-10	🇨🇳 Qwen3-Next	Alibaba	Apache 2.0, strong coding, open weights	-
2025-09-09	🌙 Kimi K2-0905	Moonshot AI	256K context, agentic coding, Modified MIT	-
2025-09-05	🇨🇳 Qwen3-Max	Alibaba	1T+ params, ranks 3rd globally, open weights	-
2025-09-03	🇪🇺 TildeOpen LLM	Tilde AI	30B params, 34 European languages, open weights	-
2025-08-21	🔬 DeepSeek-V3.1	DeepSeek	Hybrid architecture, 40% improvement, MIT license	11.4% overall
2025-08-07	🤖 GPT-5	OpenAI	Unified reasoning, multimodal, open weights	-
2025-08-05	🆓 GPT-OSS-120B	OpenAI	First open-weight since GPT-2, Apache 2.0	-
2025-08-05	🆓 GPT-OSS-20B	OpenAI	#6 Most Used - 20.7B tokens on OpenRouter, Apache 2.0	-
2025-08-01	🤗 StarCoder2	BigCode/Hugging Face	New open-source coding models, community-driven	-
2025-07-15	💻 Magistral	Mistral AI	European reasoning, open weights, self-hostable	-

*Market Share Data from OpenRouter.ai (Live Usage Stats) | 2025-09-10 | 🇨🇳 Qwen3-Next | Alibaba | Apache 2.0, strong coding | | 2025-09-09 | 🌙 Kimi K2-0905 | Moonshot AI | 256K context, agentic coding | | 2025-09-05 | 🇨🇳 Qwen3-Max | Alibaba | 1T+ params, ranks 3rd globally | | 2025-09-03 | 🇪🇺 TildeOpen LLM | Tilde AI | 30B params, 34 European languages | | 2025-08-21 | 🔬 DeepSeek-V3.1 | DeepSeek | Hybrid architecture, 40% improvement | | 2025-08-07 | 🤖 GPT-5 | OpenAI | Unified reasoning, multimodal | | 2025-08-05 | 🆓 GPT-OSS-120B | OpenAI | First open-weight since GPT-2, Apache 2.0 | - | | 2025-08-05 | 🆓 GPT-OSS-20B | OpenAI | #6 Most Used - 20.7B tokens on OpenRouter | - |

🔥 Top Performers (Real-World Usage - OpenRouter Rankings)

Rank	Model	Company	Usage	Market Position
🥇 #1	Grok Code Fast	xAI	174B tokens	50% of coding category
🥈 #2	Claude 4.5 Sonnet	Anthropic	35.5B tokens	10.2% overall usage
🥉 #3	Qwen3-Coder 30B	Alibaba	21.6B tokens	6.2% coding market
#4	Claude 4 Sonnet	Anthropic	20.8B tokens	6.0% overall
#5	GPT-OSS-20B	OpenAI	20.7B tokens	6.0% overall

Market Share by Company (OpenRouter Live Stats):

• 🥇 xAI: 26.8% (426B tokens)
• 🥈 Google: 18.9% (301B tokens)
• 🥉 OpenAI: 13.8% (220B tokens)
• 🏅 Anthropic: 13.6% (217B tokens)
• 🏅 DeepSeek: 11.4% (181B tokens)

---

📊 Model Comparison Tables

🔄 Sort by Latest Update

💡 Models sorted by most recently updated first - ⭐ indicates updates within last 30 days

🏢 Company	🤖 Model	📦 Version	📅 Release	🔄 Last Updated	💻 Coding	📊 Benchmarks	💰 Price ($/1M)	🖥️ Self-Host	🌟 Usage Rank	🔗 Link
🏢 IBM	Granite 4.0	4.0 Small	2025-10-03	2025-10-03 ⭐	✅ Good	~70% / ~75%	🆓 Free	✅ Apache 2.0	-	🔗
🇨🇳 Zhipu AI	GLM-4.6	GLM-4.6	2025-09-30	2025-09-30 ⭐	✅ Excellent	~85% / ~84%	$0.13 / $0.39	✅ Open-weight	#4 Tool Calls	🔗
🔬 DeepSeek	DeepSeek-V3.2-Exp	V3.2-Exp	2025-09-29	2025-09-30 ⭐	✅ Excellent	Experimental DSA	$0.27 / $0.41	✅ MIT	#3 Coding	🔗
🤖 Anthropic	Claude 4.5 Sonnet	Sonnet 4.5	2025-09-29	2025-09-29 ⭐	✅ Best-in-class	SWE-bench leader	$3.00 / $15.00	❌	🥈 #2 Overall	🔗
🤖 OpenAI	GPT-5 Codex	Codex	2025-09-23	2025-09-23 ⭐	✅ Best-in-class	Coding-optimized	API pricing	❌	🔗
🪶 Microsoft	Phi-4	Phi-4	2024-12	2025-09-11 ⭐	✅ Good	~70% / ~75%	🆓 Free	✅ MIT	🔗
🇨🇳 Alibaba	Qwen3-Next	Qwen3-Next	2025-09-10	2025-09-10 ⭐	✅ Good	~80% / ~84%	Varies	✅ Apache 2.0	🔗
🌙 Moonshot AI	Kimi K2-0905	K2-0905	2025-09-09	2025-09-09 ⭐	✅ Excellent	256K context	Varies	✅ Modified MIT	🔗
🇨🇳 Alibaba	Qwen3-Max	Qwen3-Max	2025-09-05	2025-09-05 ⭐	✅ Excellent	~82% / ~85%	$0.30 / $3.00	❌ API-only	🔗
🇪🇺 Tilde AI	TildeOpen LLM	30B	2025-09-03	2025-09-03 ⭐	✅ Good	EU Languages	🆓 Free	✅ Open-source	🔗
🔬 DeepSeek	DeepSeek-V3.1	V3.1	2025-08-21	2025-08-21	✅ Excellent	82%+ / 85%+	$0.27 / $0.41	✅ MIT	🔗
💻 Mistral AI	Codestral	2508	2025-08	2025-08	✅ Excellent	Coding-specialized	$0.30 / $0.90	❌	🔗
🤖 OpenAI	GPT-5	GPT-5	2025-08-07	2025-08-07	✅ Excellent	~90%+ / ~92%	$1.25 / $10.00	❌	🔗
🤖 Anthropic	Claude Opus 4.1	Opus 4.1	2025-08-05	2025-08-05	✅ Excellent	~85%+ / ~85%	$15.00 / $75.00	❌	🔗
🆓 OpenAI	GPT-OSS-120B	OSS-120B	2025-08-05	2025-08-05	✅ Excellent	91.4% / ~89%	🆓 Free	✅ Apache 2.0	🔗
🆓 OpenAI	GPT-OSS-20B	OSS-20B	2025-08-05	2025-08-05	✅ Good	~85% / 85.3%	🆓 Free	✅ Apache 2.0	🔗
🇨🇳 Alibaba	Qwen3-Coder	480B	2025-07-23	2025-07-23	✅ Excellent	Coding-optimized	🆓 Free	✅ Apache 2.0	🔗
🚀 xAI	Grok 4	Grok 4	2025-07-09	2025-07-09	✅ Excellent	~85% / ~87%	$3.00 / $15.00	❌	🔗
🇨🇳 Zhipu AI	GLM-4.5	GLM-4.5	2025-07	2025-07	✅ Good	~82% / ~82%	$0.15 / $0.45	✅ Open-weight	🔗
🌙 Moonshot AI	Kimi K2	K2	2025-07	2025-07	✅ Excellent	~85% / ~83%	Varies	✅ Modified MIT	🔗
💻 Mistral AI	Magistral	Magistral	2025-06	2025-06	✅ Good	Reasoning	Varies	❌	🔗
🔬 DeepSeek	DeepSeek-R1	R1-0528	2025-05-28	2025-05-28	✅ Excellent	81% / 85%	$0.50 / $2.15	✅ MIT	🔗

---

🏢 Sort by Company

💡 Models grouped by company/organization

🤖 OpenAI Models (7 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
GPT-5 Codex	Codex	2025-09-23	2025-09-23	✅ Best	Coding-optimized	API	❌
GPT-5	GPT-5	2025-08-07	2025-08-07	✅ Excellent	~90%+ / ~92%	$1.25 / $10	❌
🆓 GPT-OSS-120B	OSS-120B	2025-08-05	2025-08-05	✅ Excellent	91.4% / ~89%	Free	✅ Apache 2.0
🆓 GPT-OSS-20B	OSS-20B	2025-08-05	2025-08-05	✅ Good	~85% / 85.3%	Free	✅ Apache 2.0
o3	o3	2025-04	2025-04	✅ Excellent	85%+ / ~88%	$2.00 / $8	❌
o1-Pro	o1-Pro API	2025-03	2025-03	✅ Advanced	Pro reasoning	$150 / $600	❌
o3-Mini	o3-Mini	2024-12	2024-12	✅ Good	~77% / ~87%	$1.10 / $4.40	❌

🤖 Anthropic Models (3 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
Claude Sonnet 4.5	Sonnet 4.5	2025-09-29	2025-09-29	✅ Best	SWE-bench leader	$3 / $15	❌
Claude Opus 4.1	Opus 4.1	2025-08-05	2025-08-05	✅ Excellent	~85%+ / ~85%	$15 / $75	❌
Claude 3.7 Sonnet	3.7 Sonnet	2025-02-24	2025-02-24	✅ Excellent	~86% / ~84.8%	$3 / $15	❌

🔬 DeepSeek Models (4 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
DeepSeek-V3.2-Exp	V3.2-Exp	2025-09-29	2025-09-30	✅ Excellent	Experimental DSA	$0.27 / $0.41	✅ MIT
DeepSeek-V3.1	V3.1	2025-08-21	2025-08-21	✅ Excellent	82%+ / 85%+	$0.27 / $0.41	✅ MIT
DeepSeek-R1	R1-0528	2025-05-28	2025-05-28	✅ Excellent	81% / 85%	$0.50 / $2.15	✅ MIT
🆓 DeepSeek-Coder-V2	Coder-V2	2024-06	2024-06	✅ Excellent	Coding specialist	Free	✅ MIT

🇨🇳 Alibaba/Qwen Models (5 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
Qwen3-Next	Qwen3-Next	2025-09-10	2025-09-10	✅ Good	~80% / ~84%	Varies	✅ Apache 2.0
Qwen3-Max	Qwen3-Max	2025-09-05	2025-09-05	✅ Excellent	~82% / ~85%	$0.30 / $3	❌ API
🆓 Qwen3-Coder	480B	2025-07-23	2025-07-23	✅ Excellent	Coding-optimized	Free	✅ Apache 2.0
🆓 Qwen2.5-Coder	32B	2024-11	2024-11	✅ Excellent	Coding-focused	Free	✅ Apache 2.0
Qwen2.5-Max	2.5-Max	2025-01-29	2025-01-29	✅ Good	~80% / ~84%	Varies	❌ API

🇨🇳 Zhipu AI (Z.ai) Models (2 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
GLM-4.6	GLM-4.6	2025-09-30	2025-09-30	✅ Excellent	~85% / ~84%	$0.13 / $0.39	✅ Open-weight
GLM-4.5	GLM-4.5	2025-07	2025-07	✅ Good	~82% / ~82%	$0.15 / $0.45	✅ Open-weight

🌙 Moonshot AI Models (2 models)

Model	Version	Release	Last Updated	Coding	Benchmarks	Price	Self-Host
Kimi K2-0905	K2-0905	2025-09-09	2025-09-09	✅ Excellent	256K context	Varies	✅ Modified MIT
Kimi K2	K2	2025-07	2025-07	✅ Excellent	~85% / ~83%	Varies	✅ Modified MIT

Other Companies (10+ models)

🌐 Google DeepMind - Gemini 2.5 Pro, Gemini 2.5 Flash 🚀 xAI - Grok 4, Grok 4 Fast, Grok 3 🦙 Meta - Llama 4 Maverick, Llama 4 Scout 💻 Mistral AI - Codestral, Magistral, Medium 3, Pixtral Large, Large 2 🏢 IBM - Granite 4.0 🪶 Microsoft - Phi-4 🇪🇺 Tilde AI - TildeOpen LLM And more...

---

🖥️ Sort by Self-Hosting

💡 Models filtered by self-hosting capability

✅ Self-Hostable Models (20+ models)

Model	Company	Parameters	License	API Price	Last Updated
🆓 GPT-OSS-120B	OpenAI	117B (5.1B active)	Apache 2.0	Free	2025-08-05
🆓 GPT-OSS-20B	OpenAI	21B (3.6B active)	Apache 2.0	Free	2025-08-05
🆓 DeepSeek-V3.2-Exp	DeepSeek	671B (37B active)	MIT	$0.27/$0.41	2025-09-30
🆓 DeepSeek-V3.1	DeepSeek	671B (37B active)	MIT	$0.27/$0.41	2025-08-21
🆓 DeepSeek-R1	DeepSeek	671B	MIT	$0.50/$2.15	2025-05-28
🆓 DeepSeek-Coder-V2	DeepSeek	236B	MIT	Free	2024-06
🆓 Qwen3-Next	Alibaba	Various	Apache 2.0	Varies	2025-09-10
🆓 Qwen3-Coder	Alibaba	480B	Apache 2.0	Free	2025-07-23
🆓 Qwen2.5-Coder	Alibaba	32B	Apache 2.0	Free	2024-11
🆓 Kimi K2-0905	Moonshot AI	1T (32B active)	Modified MIT	Varies	2025-09-09
🆓 Kimi K2	Moonshot AI	1T (32B active)	Modified MIT	Varies	2025-07
🆓 GLM-4.6	Zhipu AI	355B MoE	Open-weight	$0.13/$0.39	2025-09-30
🆓 GLM-4.5	Zhipu AI	Various	Open-weight	$0.15/$0.45	2025-07
🆓 Llama 4 Maverick	Meta	400B	Meta License	Free	2025-04-05
🆓 Llama 4 Scout	Meta	109B	Meta License	Free	2025-04-05
🆓 Granite 4.0	IBM	8B-3B active	Apache 2.0	Free	2025-10-03
🆓 Phi-4	Microsoft	14B	MIT	Free	2025-09-11
🆓 TildeOpen LLM	Tilde AI	30B	Open-source	Free	2025-09-03
🆓 Yi-Coder	01.AI	9B / 1.5B	Apache 2.0	Free	2024-09
🆓 StarCoder2	BigCode/HF	3B-15B	BigCode	Free	2024

❌ API-Only Models (Proprietary)

Model	Company	Pricing	Performance	Last Updated
GPT-5	OpenAI	$1.25 / $10	Excellent	2025-08-07
GPT-5 Codex	OpenAI	API pricing	Best coding	2025-09-23
Claude Sonnet 4.5	Anthropic	$3 / $15	Best coding	2025-09-29
Claude Opus 4.1	Anthropic	$15 / $75	Excellent	2025-08-05
Gemini 2.5 Pro	Google	$1.25 / $10	99% HumanEval	2025-01-31
Grok 4	xAI	$3 / $15	Excellent	2025-07-09
Grok 4 Fast	xAI	$0.20 / $1.50	Cost-efficient	2025-09

---

💰 Sort by Price

💡 Models sorted by cost (cheapest first)

🆓 Free Models (Self-Hostable)

All models in the "Self-Hostable" section above are free to self-host!

💵 Budget-Friendly (< $0.50 per 1M tokens)

Model	Company	Input	Output	Total (avg)
🥇 GLM-4.6	Zhipu AI	$0.13	$0.39	$0.26
🥈 Yi-Lightning	01.AI	$0.14	$0.42	$0.28
🥉 Grok 4 Fast	xAI	$0.20	$1.50	$0.85
DeepSeek-V3.1/V3.2	DeepSeek	$0.27	$0.41	$0.34
Gemini 2.5 Flash	Google	$0.30	$2.50	$1.40
Qwen3-Max	Alibaba	$0.30	$3.00	$1.65
Codestral	Mistral AI	$0.30	$0.90	$0.60

💰 Mid-Tier ($1 - $5 per 1M tokens)

Model	Company	Input	Output	Total (avg)
GPT-5	OpenAI	$1.25	$10.00	$5.63
Gemini 2.5 Pro	Google	$1.25	$10.00	$5.63
Mistral Medium 3	Mistral AI	$1.00	$3.00	$2.00
Mistral Large 2	Mistral AI	$2.00	$6.00	$4.00
o3	OpenAI	$2.00	$8.00	$5.00
Claude Sonnet 4.5	Anthropic	$3.00	$15.00	$9.00
Grok 4	xAI	$3.00	$15.00	$9.00

💎 Premium (> $5 per 1M tokens)

Model	Company	Input	Output	Total (avg)
Claude Opus 4.1	Anthropic	$15.00	$75.00	$45.00
o1-Pro	OpenAI	$150.00	$600.00	$375.00

---

🎯 Models by Category

🏆 Frontier Models

The most advanced, cutting-edge models with state-of-the-art capabilities:

• 🥇 Grok Code Fast (xAI) - #1 Most Used - 50% of coding market, 174B tokens
• 🥈 Claude 4.5 Sonnet (Anthropic) - #2 Overall - Best coding, 35.5B tokens
• GPT-5 (OpenAI) - Unified reasoning and multimodal, $1.25/$10
• Gemini 2.5 Pro (Google) - Leading multimodal reasoning, 99% HumanEval
• Grok 4 (xAI) - First-principles reasoning, 26.8% company market share
• Qwen3-Max (Alibaba) - 1T+ parameters, ranks 3rd globally
• GPT-OSS-120B (OpenAI) - First open-weight since GPT-2, Apache 2.0

Commercial Coding Models

• Claude Sonnet 4.5 (Anthropic) - SWE-bench Verified leader
• GPT-5 Codex (OpenAI) - 7+ hour autonomous coding
• Codestral (Mistral AI) - Low-latency, fill-in-middle
• Grok 4 Fast (xAI) - Cost-efficient at $0.20/$1.50

Open-Source Coding Models

• GPT-OSS-120B (OpenAI) - 91.4% AIME, Apache 2.0
• Qwen3-Coder (Alibaba) - 480B params, autonomous coding
• DeepSeek-Coder-V2 (DeepSeek) - 236B params, MIT
• Kimi K2-0905 (Moonshot AI) - 256K context, agentic tasks
• GLM-4.6 (Zhipu AI) - Real-world coding, $0.13/$0.39
• IBM Granite 4.0 - Enterprise-ready, ISO 42001

🧠 Reasoning Models

Models that employ chain-of-thought and step-by-step problem solving:

• o3 / o1-Pro (OpenAI) - Advanced reasoning with extended thinking
• DeepSeek-R1 (DeepSeek) - Open-source reasoning champion, MIT
• Claude 3.7 Sonnet (Anthropic) - Hybrid reasoning model
• Magistral (Mistral AI) - European reasoning model
• Qwen3-Max-Thinking (Alibaba) - 100% AIME25 accuracy

🆓 Open-Source Models (2025)

Freely available models with permissive licenses (Apache 2.0, MIT, etc.) and public source code:

• GPT-OSS-120B / GPT-OSS-20B (OpenAI) — Apache 2.0; first open-weight since GPT-2; strong coding and reasoning.
• DeepSeek-R1 / V3.1 / V3.2 / Coder-V2 (DeepSeek) — MIT License; top-tier open-source reasoning and coding.
  GitHub
• Llama 4 (Scout/Maverick) (Meta) — Multimodal, open weights; supports text, image, and code tasks.
• Qwen3-Next / Qwen3-Coder / Qwen3-Max / Qwen2.5-Coder (Alibaba) — Apache 2.0; competitive coding and reasoning, multilingual.
  GitHub
• Yi-Coder (01.AI) — MIT License; 128K context, 52 programming languages, efficient for local deployment.
  GitHub
• TildeOpen LLM (Tilde AI) — European language specialist, open weights. Hugging Face
• Phi-4 (Microsoft) — MIT License; compact, efficient, strong reasoning.
• GLM-4.6 / GLM-4.5 (Zhipu AI) — Open-weight, real-world coding, multilingual.
• Kimi K2-0905 / Kimi K2 (Moonshot AI) — Modified MIT; agentic coding, 256K context.
• Codestral (Mistral AI) — Coding-specialized, low-latency, open weights.
  GitHub
• StarCoder2 (BigCode/Hugging Face) — Community-driven, open-source coding.
  GitHub
• Magistral (Mistral AI) — European reasoning, open weights.
• IBM Granite 4.0 — Apache 2.0; enterprise-ready, ISO 42001 certified.

Recent 2025 additions:

• Qwen3-Max (Alibaba) — 1T+ parameters, top-tier performance.
• GLM-4.6 (Zhipu AI) — 355B MoE, real-world coding, multilingual.
• StarCoder2 — New open-source coding models from BigCode/Hugging Face.
• Magistral — Reasoning-focused, open weights, European origin.

Most open-source models now support context windows of 128K–256K tokens, multimodal capabilities, and community benchmarks (HumanEval, SWE-bench, MMLU).

💻 Coding-Specialized Models

Optimized for software development tasks:

Commercial Coding Models

• Claude Sonnet 4.5 (Anthropic) - SWE-bench Verified leader
• GPT-5 Codex (OpenAI) - 7+ hour autonomous coding
• Codestral (Mistral AI) - Low-latency, fill-in-middle
• Grok 4 Fast (xAI) - Cost-efficient at $0.20/$1.50

Open-Source Coding Models

• GPT-OSS-120B (OpenAI) - 91.4% AIME, Apache 2.0
• Qwen3-Coder (Alibaba) - 480B params, autonomous coding
• DeepSeek-Coder-V2 (DeepSeek) - 236B params, MIT
• Kimi K2-0905 (Moonshot AI) - 256K context, agentic tasks
• GLM-4.6 (Zhipu AI) - Real-world coding, $0.13/$0.39
• IBM Granite 4.0 - Enterprise-ready, ISO 42001

🎨 Multimodal Models

Process text, images, audio, and video:

• GPT-5 (OpenAI) - Unified multimodal interface
• Gemini 2.5 Pro/Flash (Google) - Native multimodal architecture
• Claude Sonnet 4.5 (Anthropic) - Vision and document understanding
• Pixtral Large (Mistral AI) - 124B params, image understanding
• Llama 4 Maverick (Meta) - Native multimodality

🏢 Enterprise Models

Designed for business and production deployments:

• Claude Opus 4.1 (Anthropic) - ASL-3 safety, highest capability
• Gemini 2.5 Pro (Google) - Google Cloud integration
• Command A (Cohere) - Enterprise RAG, 10+ languages
• Jamba 1.6 (AI21 Labs) - Private deployment, hybrid architecture
• IBM Granite 4.0 - ISO 42001 certified, auditable

---

💻 Coding Models Deep Dive

🏆 Best Coding Models by Task

Code Generation & Autocomplete:

1. 🥇 Grok Code Fast (xAI) - #1 Most Used - 174B tokens
2. 🥈 Claude 4.5 Sonnet - Most accurate, #2 overall
3. 🥉 GPT-5 Codex - Complex algorithms

Code Review & Refactoring:

1. 🥇 Claude 4.5 Sonnet - Industry-leading, 35.5B tokens
2. 🥈 Grok Code Fast - Real-world proven, 50% market
3. 🥉 GPT-5 Codex - Large-scale refactoring

Debugging & Error Fixing:

1. 🥇 Claude 4.5 Sonnet - Clear explanations
2. 🥈 GPT-5 Codex - Deep analysis
3. 🥉 DeepSeek-V3.1 - Reasoning-based

Test Generation:

1. 🥇 Grok Code Fast - Real-world adoption leader
2. 🥈 Claude 4.5 Sonnet - Comprehensive coverage
3. 🥉 Codestral - Purpose-built for testing

Real-World Usage (OpenRouter Live Data):

• Grok Code Fast dominates with 50% of coding category
• Claude 4.5 Sonnet is #2 most deployed model
• Qwen3-Coder 30B is #3 in coding workflows

🏆 Coding Benchmarks (SWE-bench Verified)

1. 🥇 Claude Sonnet 4.5 - State-of-the-art
2. 🥈 GPT-5 Codex - 7+ hour autonomous coding
3. 🥉 GPT-OSS-120B - 91.4% AIME, open-source leader
4. Kimi K2-0905 - Agentic coding excellence
5. Qwen3-Coder - 480B autonomous generation
6. DeepSeek-V3.2-Exp - Sparse attention efficiency
7. GLM-4.6 - Real-world tasks
8. Grok 4 Fast - Best cost/performance
9. IBM Granite 4.0 - Enterprise-grade
10. Phi-4 - Compact reasoning

📏 Context Window Comparison

For working with large codebases:

1. 🌙 Kimi K2-0905 - 256K tokens (≈192K lines)
2. 🇨🇳 GLM-4.6 - 200K tokens (≈150K lines)
3. 🤖 Claude Sonnet 4.5 - 200K tokens (≈150K lines)
4. 🇨🇳 Qwen3-Coder - 128K tokens (≈96K lines)
5. 🔬 DeepSeek-Coder-V2 - 128K tokens (≈96K lines)

---

🔗 Official Resources

🤖 OpenAI

• Website: openai.com
• API Docs: platform.openai.com
• Models: GPT-5, GPT-5 Codex, GPT-OSS-120B/20B, o3, o1-Pro

🤖 Anthropic

• Website: anthropic.com
• API Docs: docs.anthropic.com
• Models: Claude Sonnet 4.5, Claude Opus 4.1, Claude 3.7 Sonnet

🌐 Google DeepMind

• Website: deepmind.google/gemini
• API Docs: ai.google.dev
• Models: Gemini 2.5 Pro, Gemini 2.5 Flash

🚀 xAI

• Website: x.ai
• Chat: grok.x.ai
• Models: Grok 4, Grok 4 Fast, Grok 3

🔬 DeepSeek

• Website: deepseek.com
• GitHub: github.com/deepseek-ai
• Models: DeepSeek-R1, V3.1, V3.2-Exp, Coder-V2

🇨🇳 Alibaba/Qwen

• Website: qwenlm.github.io
• GitHub: github.com/QwenLM
• Models: Qwen3-Max, Qwen3-Next, Qwen3-Coder

🇨🇳 Zhipu AI (Z.ai)

• Website: chatglm.cn
• GitHub: github.com/THUDM
• Models: GLM-4.6, GLM-4.5

🌙 Moonshot AI

• Website: moonshot.cn
• Chat: kimi.moonshot.cn
• Models: Kimi K2-0905, Kimi K2

🦙 Meta AI

• Website: llama.meta.com
• GitHub: github.com/meta-llama
• Models: Llama 4 Maverick, Llama 4 Scout

💻 Mistral AI

• Website: mistral.ai
• Chat: chat.mistral.ai
• Models: Codestral, Magistral, Mistral Large 2, Pixtral Large

🏢 IBM Research

• Website: ibm.com/granite
• GitHub: github.com/ibm-granite
• Models: Granite 4.0

🪶 Microsoft Research

• Website: microsoft.com/research
• GitHub: github.com/microsoft
• Models: Phi-4

🇪🇺 Tilde AI

• Website: tilde.ai
• Hugging Face: huggingface.co/tilde-research
• Models: TildeOpen LLM

---

📈 Performance Benchmarks

🏆 Coding Benchmarks (SWE-bench Verified)

1. 🥇 Claude Sonnet 4.5 - State-of-the-art
2. 🥈 GPT-5 Codex - 7+ hour autonomous coding
3. 🥉 GPT-OSS-120B - 91.4% AIME, open-source leader
4. Kimi K2-0905 - Agentic coding excellence
5. Qwen3-Coder - 480B autonomous generation
6. DeepSeek-V3.2-Exp - Sparse attention efficiency
7. GLM-4.6 - Real-world tasks
8. Grok 4 Fast - Best cost/performance
9. IBM Granite 4.0 - Enterprise-grade
10. Phi-4 - Compact reasoning

🧮 Reasoning Benchmarks (AIME 2025)

1. 🥇 Qwen3-Max-Thinking - 100% accuracy
2. 🥈 o1-Pro - Advanced reasoning
3. 🥉 GPT-OSS-120B - 91.4%
4. GPT-5 - ~90%
5. DeepSeek-R1 - 81%
6. Grok 4 - ~85%

📚 General Knowledge (MMLU)

1. 🥇 GPT-5 - ~92%
2. 🥈 o3 - ~88%
3. 🥉 Gemini 2.5 Pro - 86.4%
4. DeepSeek-V3.1 - 85%+
5. Claude Opus 4.1 - ~85%

---

💰 Cost Analysis

🆓 Most Affordable Options

Free Self-Hostable (20+ models):

• All models in "Self-Hostable" section can be run for FREE!
• No API costs when self-hosting
• Just need appropriate hardware

Cheapest Commercial APIs:

1. 🥇 GLM-4.6 - $0.13/$0.39 per 1M tokens
2. 🥈 Yi-Lightning - $0.14/$0.42 per 1M tokens
3. 🥉 Grok 4 Fast - $0.20/$1.50 per 1M tokens

💎 Best Value (Performance/Cost)

1. DeepSeek-V3.1 - Excellent performance, ultra-low cost
2. Grok 4 Fast - Frontier performance, $0.20/$1.50
3. Qwen3-Max - Top-3 global, $0.30/$3.00
4. Gemini 2.5 Flash - Fast & capable, $0.30/$2.50
5. Claude Sonnet 4.5 - Best coding, $3/$15 (reasonable)

---

🖥️ Self-Hosting Guide

🔧 Hardware Requirements

Small Models (7-20B parameters):

• GPU: 1x RTX 3090/4090 (24GB VRAM)
• RAM: 32GB+
• Storage: 50GB+ SSD
• Examples: Phi-4, GPT-OSS-20B, Yi-Coder

Medium Models (30-70B parameters):

• GPU: 2x RTX 4090 or 1x A100 (40-80GB)
• RAM: 64GB+
• Storage: 100GB+ SSD
• Examples: Qwen3-Coder, TildeOpen LLM, Llama 4 Scout

Large Models (120B+ parameters):

• GPU: 4x A100 (80GB) or 8x RTX 4090
• RAM: 128GB+
• Storage: 200GB+ SSD
• Examples: GPT-OSS-120B, Pixtral Large

📦 Deployment Options

Local Inference:

• Ollama - Easy local deployment (ollama.ai)
• LM Studio - User-friendly GUI (lmstudio.ai)
• llama.cpp - Efficient C++ implementation
• vLLM - High-throughput serving (vllm.ai)

Cloud Self-Hosting:

• Hugging Face Inference - Managed deployment
• AWS/GCP/Azure - Full control, scalable
• RunPod/Vast.ai - GPU rental platforms
• Modal/Replicate - Serverless options

🖥️ Self-Hosted Models (2025)

Models that can be run locally or on your own infrastructure (open weights, permissive license) with public source code:

Coding:

• DeepSeek-Coder-V2 (MIT) — GitHub
• Qwen3-Coder (Apache 2.0) — GitHub +- GPT-OSS-120B (Apache 2.0)
• Codestral (Mistral AI, open weights) — GitHub
• StarCoder2 (BigCode/Hugging Face) — GitHub

General Use:

• GPT-OSS-20B (Apache 2.0)
• Phi-4 (MIT) — GitHub
• DeepSeek-V3.1 (MIT) — GitHub
• Llama 4 Scout (Meta license) — GitHub
• GLM-4.6 (Zhipu AI, open weights) — GitHub

Enterprise:

• IBM Granite 4.0 (Apache 2.0, ISO 42001)
• Llama 4 Maverick (Meta license) — GitHub
• Qwen3-Max (Apache 2.0) — GitHub
• Magistral (Mistral AI, open weights)

Recent 2025 additions:

• Qwen3-Max (Alibaba) — 1T+ parameters, self-hostable.
• GLM-4.6 (Zhipu AI) — 355B MoE, open weights, self-hostable.
• StarCoder2 — New open-source coding models, easy to self-host.
• Magistral — Reasoning, open weights, European origin.

Most self-hosted models now support large context windows, multimodal tasks, and efficient deployment on consumer or enterprise hardware.

---

🔍 Model Selection Guide

💼 By Use Case

🏢 Enterprise Production:

• Best: Claude Opus 4.1, IBM Granite 4.0
• Budget: DeepSeek-V3.1, Qwen3-Max
• Self-Host: IBM Granite 4.0, Llama 4

💻 Software Development:

• Best: Claude Sonnet 4.5, GPT-5 Codex
• Popular: Grok Code Fast (50% market share)
• Self-Host: DeepSeek-Coder-V2, Qwen3-Coder

🔬 Research & Analysis:

• Reasoning: o1-Pro, Qwen3-Max-Thinking
• General: GPT-5, Gemini 2.5 Pro
• Open: GPT-OSS-120B, DeepSeek-R1

🎨 Creative & Multimodal:

• Best: GPT-5, Gemini 2.5 Pro
• Vision: Pixtral Large, Claude Sonnet 4.5
• Affordable: Gemini 2.5 Flash

🌍 Multilingual & Regional:

• Chinese: Qwen3-Max, GLM-4.6, Kimi K2
• European: TildeOpen LLM (34 languages)
• Global: GPT-5, Gemini 2.5 Pro

💰 By Budget

🆓 Free (Self-Host Only):

• GPT-OSS-120B/20B, DeepSeek models, Llama 4, Phi-4

💵 Budget-Friendly ($0.10-$0.50 per 1M):

• GLM-4.6, Yi-Lightning, Grok 4 Fast, DeepSeek-V3.1

💎 Mid-Range ($1-$5 per 1M):

• GPT-5, Gemini 2.5 Flash, Claude Sonnet 4.5, Grok 4

👑 Premium ($10+ per 1M):

• Claude Opus 4.1, o1-Pro, Gemini 2.5 Pro

⚡ By Performance Needs

Speed Priority:

• Gemini 2.5 Flash, Grok 4 Fast, Claude 3.7 Sonnet

Quality Priority:

• Claude Opus 4.1, GPT-5, o1-Pro

Balanced:

• Claude Sonnet 4.5, GPT-5, DeepSeek-V3.1

---

🎯 Awesome AI IDEs

Awesome AI IDEs - A comprehensive list of AI-powered IDEs with categories and details.

---

🤝 Contributing

We welcome contributions from the community! Here's how you can help:

📝 How to Contribute

1. Fork the Repository
2. Add/Update Model Information
   • Verify from official sources
   • Include benchmark data
   • Add pricing information
   • Provide official links
3. Follow the Style Guide
   • Use emojis consistently
   • Match existing formatting
   • Keep tables aligned
4. Submit a Pull Request
   • Describe your changes
   • Include sources/verification
   • Follow UPDATE_RULES.md

🎯 What We Need

• ✅ New model releases and updates
• ✅ Benchmark results and performance data
• ✅ Pricing updates and corrections
• ✅ Bug fixes and typo corrections
• ✅ Additional resources and links
• ✅ Real-world usage insights
• ✅ Community feedback and reviews

📋 Guidelines

Please read our detailed contribution guidelines:

• CONTRIBUTING.md - Full contribution guide
• UPDATE_RULES.md - Update standards

🙏 Contributors

Special thanks to all contributors who help keep this matrix up-to-date!

---

📜 License

Project License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).

Copyright (c) 2025 ReadyPixels

For full license terms, see: GNU GPL v3.0

📊 Data Sources Attribution

This project aggregates information from:

• OpenRouter Rankings - openrouter.ai - Real-world usage statistics
• LLM-Stats.com - llm-stats.com - Benchmark aggregation
• Official Documentation - Direct from model providers
• Community Contributions - GitHub contributors

⚖️ Model Licenses

Each AI model listed has its own license. Please refer to official documentation:

• Open Source: Apache 2.0, MIT, Modified MIT, etc.
• Proprietary: API usage terms from providers
• Enterprise: Contact providers for licensing

---

🔗 Quick Links

• 📊 OpenRouter Rankings - Live usage data
• 📈 LLM-Stats.com - Benchmark comparisons
• 🤗 Hugging Face - Model repository
• 📖 Awesome Lists - Curated lists

---

⭐ Star this repository if you find it helpful! ⭐

Last Updated: October 7, 2025

---

📚 Additional Resources

• OpenRouter Rankings - openrouter.ai/rankings
• LLM-Stats - llm-stats.com
• HuggingFace Leaderboard - huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
• Awesome LLM - github.com/Hannibal046/Awesome-LLM

---

🚀 The most comprehensive AI models matrix on the internet! Made with ❤️ by ReadyPixels for the global AI community Last Updated: October 7, 2025